Guide
Omics

One code, many states

What each layer above the genome records, why some layers are so much harder to measure than others, and what it takes to turn one measurement into a result someone can act on

A joint guide

+ Guide · Omics

One code, many states.

Each omics layer records a different part of what cells are doing with one genome. The guide runs from the chemistry that makes DNA and RNA easy to measure, and proteins and metabolites hard, through the instruments and the chain from blood tube to signed report, to the evidence a test needs and the rules it meets in the United States, the European Union and India.

+ The central question

Every cell in the body carries the same genome. What does each omics layer record that the genome cannot, why are some layers so much harder to measure than others, and what does it take to turn one of those measurements into a result someone can act on?

Starts at:the central dogma at school-biology level. No omics, statistics or instrument background assumed.Ends at:reading an omics result critically, and knowing how the test behind it is proven to measure what it claims and how that test is regulated in the United States, the European Union and India. Built around:one pair of identical twins, one of whom spent 340 days in orbit: the same genome in both. NASA counts ten research teams in its Twins Study, whose paper covers 25 months.

The central question of this guide follows from that episode: Every cell in the body carries the same genome. What does each omics layer record that the genome cannot, why are some layers so much harder to measure than others, and what does it take to turn one of those measurements into a result someone can act on?

Three words in that question are used in a fixed sense. The genome is the complete DNA sequence a person inherits, the same in nearly every cell of the body. Omics measures a whole set of molecules of one kind in a single experiment, and each such set is a layer: the genome first, then the sets of molecules a cell makes as it uses its genes. In the last clause, a result is the answer a test gives for one sample, and someone acts on it when a doctor, a patient or a manufacturer decides something because of that answer.

In brief, the genome fixes what a cell could become, and the other layers record what it has become and what it is doing. The epigenome is the set of chemical marks on DNA and on the proteins that package it, and it records which genes a cell keeps open for use, often through many divisions. The transcriptome is the cell's RNA, and it records which genes the cell is using now and how heavily. The proteome is the full set of proteins, and it records which ones the cell has made and in what modified forms. The metabolome is the full set of small molecules, such as sugars, amino acids and fats, and it records the chemistry those proteins carry out. Further layers record which cells a sample holds and where each sat in its tissue, which microbes share the body, and what cells release into the blood. Layers further from the genome show more directly what the body is doing at the time of sampling.

The two things confused in 2018, a gene's sequence and its expression, are both read from nucleic acids: the sequence from DNA, and expression usually from RNA. The layers read from nucleic acids are the easiest to measure, for two chemical reasons that Chapter 2 sets out. The first is copying: enzymes copy DNA, and RNA once it has been written into DNA, so a few molecules can be multiplied until an instrument detects them. The second is pairing: each strand binds a partner strand letter by letter, so a probe for any sequence can be designed from the sequence alone.

Proteins and metabolites offer neither property. No enzyme copies a protein or a metabolite, and no rule like the pairing of DNA strands turns a protein's sequence into the sequence of a binder for it. An instrument therefore has to detect the molecules the sample already holds, and tell them apart by their mass or by a binder that has been raised, selected or designed for each one and then tested. For that reason the layers that show most directly what the body is doing are the hardest to measure.

The third clause concerns everything after the measurement. An omics result reaches a decision through a chain of steps: a specimen drawn and handled, an assay and an instrument, software that turns signals into numbers, a person who interprets them and a report that someone signs. A fault at any step can hide a real difference or make noise look like a finding. Before anyone relies on the result, evidence has to show that the test measures what it claims, and separate evidence has to show whether acting on its answer helps the people tested (Chapters 19 and 22). The law of each jurisdiction sets which parts of that evidence a regulator reviews, and when.

Parts I to V answer the first clause, and Part I also answers the second. Part I sets the genome against the state of a cell. Its Chapter 2 explains the chemistry that separates the easy layers from the hard ones, and Chapter 3 the few physical properties that instruments read. Parts II to IV take the layers in turn: the nucleic-acid layers, read by copying and pairing; proteins and metabolites, which allow neither; and single cells, tissue maps, microbes and the material that cells shed into blood. Each of those chapters shows the second clause at work, because its methods follow from what its molecule allows. Part V reads the Twins Study with every layer in hand.

Parts VI to IX answer the third clause. Part VI treats an omics test as a designed system: the instrument, the chain from tube to report, the reading of a result, the evidence that the test works, and what keeps the test proven once people rely on it. Part VII shows where omics already informs decisions. It then sets out what a signature has to show before a test can be built on it, a signature being a pattern across many measurements that separates one group of people from another. Part VIII describes the makers and the frontier as of October 2026, and Part IX how the United States, the European Union and India regulate omics tests at the same date.

Three kinds of reader can take shorter routes. An engineer new to the field can read Chapters 1 to 3 for the model and then one method of each kind: sequencing in Chapter 4, mass spectrometry in Chapter 8 and binding in Chapter 9. Chapters 16 and 17 then show the instrument as a system and the chain from tube to report. Chapter 19 covers the life cycle of the software that measures, and Chapter 20 how a test stays proven when a supplier or the design changes or a complaint reveals a fault.

A founder or product lead weighing whether an omics measurement can carry a product should read Chapters 1, 2 and 15 for what the layers record and why some cost more to measure. Chapters 18, 21 and 22 then show what a finding is worth and what a test needs. Chapter 19 sets out what the team behind a test must show before the first result, and Chapter 20 what keeps the test proven once people rely on it. The chapter of Part IX for the first market sets out which route the test takes: Chapter 25 for the United States, Chapter 26 for the European Union or Chapter 27 for India. Chapters 23 and 24 show who builds the tools and what is changing.

A regulatory or quality professional who will own the file for a test should begin with Chapters 17, 19 and 20 on the chain, the evidence before release and the work that keeps a test proven after it. Chapter 22 then sets out the ladder of evidence, from a correct measurement to a proven benefit, and Chapters 25 to 27 the three jurisdictions. Chapter 23 lists the instruments, kits and tests of each layer with the regulator's decisions they hold, and the chapter on the layer the test reads supplies the method's limits.

How to read this guide

The guide has 27 chapters in nine parts. A short paragraph at the head of each part says what its chapters do, and each chapter starts from a question that the one before it leaves open. Cross-references give chapter numbers, so a case told in full in one chapter is only pointed to in the others, and the opening story returns in the closing pages of Chapter 27. The plates carry a numbering of their own, apart from the chapters, and any plate can be cited by its number alone. Parts VIII and IX describe the field and its regulation as they stood in October 2026. Four kinds of box recur, shown here as they appear in the chapters:

The mathematics goes no further than doubling, powers of ten, percentages, ratios and squares, and a logarithmic scale is described as a count of tenfold steps or of doublings. Each chapter works one number through in a solved example, with real values, every step shown and its assumptions stated. A plate that is a schematic rather than a drawing of data says so.

Inside the plates, color is a legend and never decoration. DNA, RNA, protein, metabolite and chemical modification each have one color throughout the guide, the measured readout is always blue, and one orange mark in each plate points at the detail that matters most. A key strip under every plate lists only the colors that plate uses.

The color key

+ Part I · Foundations

What the genome cannot tell.

At typical levels in the plasma of a healthy adult, each molecule of the signaling protein interleukin-6 is outnumbered by about 2.5 billion molecules of albumin, and unlike a molecule of DNA, the rare protein cannot be copied to make it easier to find. This part builds the model that the later chapters rely on: why one genome can produce cells as different as a neuron and a liver cell, why DNA and RNA can be copied and recognized by pairing while proteins and metabolites cannot, and which few physical properties omics instruments read. Nothing in it assumes more than school biology.

+ The questionIf a neuron and a liver cell carry identical DNA, what makes them different?

The instruction set and the state

A human cell carries two copies of its genome, one inherited from each parent, and each copy is about 3.1 billion base pairs long. Somewhere around 20,000 stretches of that sequence are genes that encode proteins. Their protein-coding portions occupy less than 1.5 % of the total, and the rest includes the switches, spacers and structural sequence that control when and where those genes are used. With a handful of exceptions, every nucleated cell in the body carries essentially the same sequence. The exceptions are instructive rather than fatal to the rule: red blood cells discard their nucleus, immune cells cut and rejoin their receptor genes, eggs and sperm carry one copy instead of two, and every dividing cell slowly accumulates a few new mutations. None of these explains why a neuron is a neuron.

Direct evidence that a specialized cell can keep the information for other cell types came from frogs. At the Department of Zoology in Oxford, John Gurdon took nuclei from the cells lining the intestine of feeding tadpoles of Xenopus laevis and transplanted each one into an egg whose own nucleus had been destroyed by irradiation. Gurdon began the work as a graduate student of Michail Fischberg and published it in 1962. The donor cells were visibly specialized, with what his paper called "the striated border typical of cells having an absorptive function". Of 726 eggs in the first round of transfers, 10 grew into normal feeding tadpoles, and a genetic marker proved that those tadpoles had developed from the transplanted nucleus.

Ten tadpoles from 726 transfers is a small yield, and Gurdon stated his conclusion with care: the differentiation of a cell "cannot be dependent upon the incapacity of its nucleus to give rise to other types of differentiated cell". A nucleus taken from a cell built to absorb food still held what was needed to build a swimming, feeding tadpole, so the donor had kept the information for the whole animal, and what made it an intestinal cell was not written in its sequence. The experiment could not show where the difference was recorded, and each layer that omics measures is one place to look for it.

What differs between cells is which parts of the genome they use. A neuron is long, electrically excitable and packed with ion channels, while a liver cell is compact, stuffed with enzymes that detoxify drugs and make blood proteins, and has no use for an axon. Biologists count a few hundred distinct cell types in the human body, and every one of them is a different selection from the same genome. The genome is the instruction set: it constrains what a cell can become. What the cell actually is at a given moment is its state: which instructions it is using, how intensely, and with what result. Two cells with identical genomes can be in radically different states, and one cell can change state within minutes.

01 — Same code, different cells

DNARNAProteinFocal detail

Reading is regulated at every step

The route from instruction to function has several steps, and the cell controls each one. DNA is packed around protein spools called nucleosomes, and chemical marks on the DNA and on the spools decide whether a gene is accessible at all. An accessible gene can be copied into RNA, a process called transcription, at a rate the cell sets. The RNA is edited, exported and eventually destroyed, and its lifetime is regulated too. A messenger RNA is read by a ribosome to build a protein, a process called translation, again at a controllable rate.

Biologists say that a gene is expressed when it is transcribed, and they usually measure its expression as the amount of its RNA, or of the protein made from it. The finished protein may be cut, folded, decorated with phosphate groups or sugars, sent out of the cell, or degraded. Many proteins are enzymes, and their activity determines which small molecules, the metabolites, are made and consumed.

Every one of those control points is a place where information enters that was not present in the step before. An accessible gene may be transcribed slowly or fast, so knowing that it is accessible does not reveal how much RNA is being made from it. The RNA level does not reveal how much protein exists, because translation rates and protein lifetimes vary by orders of magnitude between genes. The amount of an enzyme does not reveal how much chemistry it is doing either, because enzymes are switched on and off by modification and by the metabolites around them. These control points give the first answer to the central question. Each layer downstream of the genome records a decision the cell has made, and that decision is not written in the sequence.

Omics measures a whole set at once

The suffix -ome means the complete set: the genome is all of an organism's DNA sequence, and the proteome all of its proteins. Omics is the practice of measuring an entire set at once instead of one member at a time. An assay is a laboratory test for a substance in a sample, and a classical assay measures one substance at a time, so it asks a narrow question, such as how much of one protein a sample holds. An omics experiment asks a wide one, such as which of the thousands of proteins in a sample differ between two groups.

The width is both the strength and the cost. Measuring everything at once lets a study find molecules that no one knew to look for. A biomarker is a molecule whose level indicates a state of the body, such as a disease or a response to treatment, and many candidate biomarkers are now first found this way. But an instrument that measures twenty thousand things in one run cannot optimize itself for any one of them. Each individual measurement is typically less precise than a dedicated assay would be, and with twenty thousand measurements some will look interesting by chance alone. Both costs return later in this guide, the first in the chapters on individual layers and the second in Chapter 18.

02 — The omics stack

DNARNAProteinMetaboliteModificationFocal detail

Each layer changes at its own rate

The layers also differ in how fast they change, and that matters as much as what they record. The inherited sequence is fixed for life. Epigenetic marks range from modifications of the spool proteins, the histones, that can change within minutes to DNA methylation patterns that persist for years, and many are copied through cell division. Messenger RNAs in mammalian cells have lifetimes measured in hours, and proteins last longer. In a 2011 study at the Max Delbrück Center in Berlin, Schwanhäusser and colleagues labeled newly made RNA and protein in one line of mouse fibroblasts, a kind of connective-tissue cell, and measured the turnover of both for more than 5,000 genes. They found a median half-life, the time in which half of the molecules present are destroyed, of about 9 hours for messenger RNA and 46 hours, about two days, for protein. Many metabolites turn over in seconds to minutes.

Each layer therefore records a different span of time, much as a camera records a different interval at each shutter speed. The transcriptome reflects the last few hours of the cell's decisions, and the metabolome the last few minutes of its chemistry. That difference has an immediate practical consequence, because the faster a layer changes, the more it can change after the sample is taken. A blood tube left on a bench keeps metabolizing, and cells under stress during tissue collection start changing their RNA before anyone measures it. The further a layer sits from the genome, the more the handling of the sample becomes part of the result.

Twins as an experiment with the genome held constant

The design of the NASA Twins Study rests on the distinction between instruction set and state. One of a pair of identical twins spent 340 days aboard the International Space Station while his brother stayed on Earth, and research teams measured most of the molecular layers this guide covers in both men. Identical twins arise from a single fertilized egg, so the brother on the ground served as a genetically matched control. The pair is close to an ideal experiment for this guide's question, because the inherited genome is held constant. A difference between the twins, or between one twin before and after flight, therefore points to another layer, unless it is one of the few changes in DNA that cells acquire during life. Chapter 15 sets out the study's design and those acquired changes, and reads its results layer by layer once the guide has built the tools to interpret them.

The 7 % figure at the center of the opening came from this study's preliminary results. The inherited sequence had not changed: what had shifted, and in part stayed shifted, was the expression of a set of genes. Reading that change in state as a change in the code is the confusion this chapter exists to prevent.

+ The questionIf a whole genome can now be read for a few hundred dollars, why can a drop of blood not be measured for every protein?

Signal per molecule is the whole problem

A sequencer can read a person's whole genome from a blood sample, yet no method measures every protein in that same blood as completely and as cheaply as sequencing reads the genome, and most of the difference lies in the molecules themselves. Every detector, whether a camera, a photomultiplier or an ion counter, needs a minimum amount of signal before it can report anything above its own noise. One molecule on its own produces very little: a few photons if it carries a fluorescent dye, or one ion if it survives the trip through a mass spectrometer. The central engineering question of every omics measurement is therefore the same. How can an instrument get enough signal from each molecule of interest, and how can it tell that the signal came from that molecule and not from something else?

There are only a few ways to answer that question. A method can make more copies of the target until the signal is large. It can attach a label, a tag such as a fluorescent dye that is easy to detect, to a probe or binder that recognizes the target and nothing else, and make that label bright. Or it can use a detector sensitive enough to see individual molecules and count them one at a time. Nucleic acids happen to allow the first two answers cheaply and almost universally. That single fact explains most of the difference between the layers of the omics stack in cost, in breadth and in depth. Breadth is how many different molecules a method reaches, and depth is how scarce a molecule it can still detect.

Nucleic acids can be copied, and proteins cannot

DNA is a template for itself. A DNA polymerase moves along one strand and builds its partner, adding the base that pairs with each one it reads. Two primers, short pieces of DNA that pair with the sequence on either side of the target, give the polymerase its starting points. When a reaction is supplied with the two primers, the four building blocks and heat cycling to separate the strands, each cycle can at most double the number of copies. The method is called the polymerase chain reaction, or PCR. Thirty cycles of perfect doubling would turn one starting molecule into about a billion, because 2 multiplied by itself 30 times is about 1.07 billion. Real reactions do not copy every molecule in each cycle, and they eventually level off, so a billion is the ceiling. Even so, a signal that was invisible becomes large enough to measure with ordinary optics.

RNA can join in with one extra step. An enzyme called reverse transcriptase copies RNA into DNA, called complementary DNA or cDNA, and from there the DNA machinery takes over, which is why transcriptomics runs on the same instruments as genomics. Proteins cannot be copied. Information in the cell flows from nucleic acid to protein and never back, so there is no enzyme that reads a protein and builds another one like it. Metabolites cannot be copied either, because they are made by chains of enzymes, not from templates. The proteins and metabolites in the tube are all that any measurement of them will ever have to work with.

03 — Two tricks only nucleic acids allow

DNAProteinFocal detail

Nucleic acids can be recognized by design

The second trick, pairing, is just as important and less often noticed. Because A pairs with T and G pairs with C, a probe for any nucleic-acid sequence can be written from the sequence alone. If a target reads ATGCCGTA, its partner is known before anything is made: it can be synthesized chemically within days, and its binding strength can be predicted from its composition. The same chemistry works for every target. Laboratories run probes or assays in sets designed to be used together, called panels, and a panel of ten thousand DNA probes amounts to ten thousand entries in a design file.

Proteins are recognized by shape, not by a code. A protein-binding reagent such as an antibody has to be raised, selected or designed for each target, one at a time, because no rule like base pairing turns a protein's sequence into the sequence of its binder. Each reagent must then be tested to show that it binds the right protein and not a similar one. A panel of five thousand proteins needs five thousand validated reagents, and ten thousand if the method uses two binders per protein. Every one of those reagents can fail in its own way, and its specificity has to be proved, not predicted.

The range of abundance in plasma

A second difference compounds the first. Nearly every nuclear gene is present at two copies per cell, one from each parent, so nearly every genomic target starts at the same abundance, the number of copies of a molecule in the sample. The genome is flat, and blood plasma is the opposite. Albumin circulates at around 40 mg/mL, while signaling proteins such as interleukin-6 sit at a few picograms per milliliter in a healthy person. That is a span of about ten orders of magnitude, and the rare proteins are often the informative ones. The span from the most to the least abundant molecule in a sample is its dynamic range, and an instrument has a dynamic range of its own: the span between the largest and the smallest signal it can record at once.

A span of about ten orders of magnitude in the sample decides what an instrument that measures everything at once can see. A mass spectrometer receives ions in proportion to what is in the sample, and it has a limited capacity per second and a limited dynamic range in one scan. If albumin and a few dozen other abundant proteins take almost all of that capacity, the rare proteins are never counted at all. That is why plasma proteomics spends so much effort removing or damping the abundant proteins before measurement, and why it still reaches far fewer proteins per sample than a sequencer reaches genes.

04 — Ten orders of magnitude in one tube

ProteinFocal detail

Copying, pairing and abundance predict the order of difficulty

Copying, pairing and dynamic range together predict a ranking of the layers that matches practice. The genome is the easiest layer, because it can be copied and paired and every target starts at the same abundance. The transcriptome is next: it can be copied after reverse transcription and paired, but its molecules span a wider range of abundance and change quickly. The proteome is harder, because nothing in it copies or pairs, its abundances span ten orders of magnitude, and each protein comes in modified forms. The metabolome is the hardest to identify, because nothing copies or pairs, its molecules are chemically diverse, and there is no template from which to predict what should be there.

The ranking also explains why so many methods for proteins and other molecules work by converting them into DNA. If two antibodies that recognize a protein each carry a short DNA tag, the tags can be joined and then copied and counted like any other DNA. A method of that kind borrows the two tricks for a molecule that does not have them, and Chapter 9 describes one in detail.

+ The questionHow is a molecule detected at all if it can be neither copied nor paired?

Every instrument measures a proxy

At the moment a sequencer detects a base being added to a growing strand of DNA, what it usually records is a flash of colored light, and the base itself is inferred from the color. No omics instrument observes biology directly. Each one converts some physical property of a molecule into a signal that a detector can count, whether photons, ions or an electrical current, and the biology is inferred from that signal through a chain of conversions that Chapter 16 follows in detail. Chapter 2 showed that only nucleic acids can be copied and paired, so a protein or a metabolite has to be detected through one of its physical properties without being copied first. In practice, the field relies mostly on five physical properties, and almost every method in this guide reads one of them or a combination of them.

05 — From molecule to count

DNAProteinReadoutFocal detail

The five are the order of the bases in a nucleic acid, the mass of a molecule, the shape that lets a reagent bind it, the light it emits or scatters, and the speed at which it travels through a medium. Each has a characteristic strength and a characteristic blind spot, so the property a method reads indicates most of what is needed to judge the method. The main exception to the five is nuclear magnetic resonance, or NMR, which reads the spin of atomic nuclei. In a strong magnetic field certain nuclei absorb radio waves at frequencies set by their chemical surroundings, so an NMR spectrum reports the atomic environments in a molecule, and Chapter 10 describes how metabolomics uses it.

Sequence: reading order by pairing

Sequencing reads the order of bases in DNA, or in RNA converted to DNA. Most instruments do it by making a partner strand one base at a time and detecting which base was added, usually as the flash of colored light described at the start of this chapter. Others pull a single strand through a nanometer-scale pore and read the change in electrical current as each group of bases passes. Either way, sequencing depends on the pairing rule that makes nucleic acids special, which is why it is the dominant readout for the genome, the transcriptome and, after chemical conversion, the epigenome. The sequence that an instrument reports from one molecule is called a read, and Chapters 4 to 7 cover how reads are made and used.

Weigh: mass spectrometry

A mass spectrometer gives molecules an electric charge, moves them through electric or magnetic fields, and measures how they respond. The response depends on the ratio of mass to charge, written m/z, which high-resolution instruments measure to within a few thousandths of a unit. Mass is universal: any molecule that can be charged and brought into a vacuum can be weighed, and the instrument does not need to know in advance what to look for.

That universality has two costs. The instrument sees whatever is abundant first, so rare molecules compete with common ones for its capacity, as Chapter 2 showed. Mass alone also rarely identifies a molecule, because many different molecules share a mass, so identity usually comes from breaking the molecule and weighing the pieces as well. Mass spectrometry is the main readout for proteins and metabolites. Chapters 8 and 10 cover its use, and Chapter 23 sets out who makes the instruments as of October 2026.

Bind: affinity reagents

A binder, usually an antibody or a short folded nucleic acid called an aptamer, recognizes its target by shape. The binder carries a reporter: an enzyme that produces color, a fluorescent dye, a metal atom or a DNA tag. Because the reporter can be amplified after binding, affinity methods can reach very low concentrations. An enzyme reporter turns over many substrate molecules for each bound target, and thirty cycles of PCR can copy a DNA tag up to about a billion-fold, as Chapter 2 showed.

The limit is that a binder finds only what it was made for. An affinity method measures a predefined list of targets, and each measurement is only as specific as its reagent. A binder that also sticks to a related protein reports both, and nothing in the signal reveals the mistake. Chapter 9 covers affinity proteomics, and Chapters 11 and 14 rely on binders to label cells and vesicles.

Light and travel time: counting objects and separating mixtures

Light is the reporter behind most of the other methods, but it becomes a measurement principle of its own when an instrument looks at individual objects. A flow cytometer measures each cell as it passes a laser, a microscope images each molecule in a tissue section, and the most sensitive instruments detect single fluorescent molecules. When objects can be seen one at a time, the measurement changes character, because instead of estimating how much signal a mixture gives, the instrument counts events, each event being one object detected. Counting is more robust than estimating intensity, and it is the basis of the digital methods that recur throughout this guide.

Separation methods do not identify anything on their own. Liquid chromatography pushes a mixture through a packed column, and molecules emerge at different times depending on how strongly they stick to the packing. Gas chromatography does the same for volatile molecules, and electrophoresis sorts by size and charge in an electric field. The job of separation is to reduce complexity, because a detector presented with a few molecules at a time performs far better than one presented with thousands at once. The time at which a molecule emerges, its retention time, also becomes an extra coordinate for identifying it.

06 — Five ways to measure a molecule

DNARNAProteinMetaboliteModificationFocal detail

Real methods combine the principles

Almost every method in practical use chains two or more of these principles, and the combination matters because each principle covers the blind spot of another. Separation spares a mass spectrometer the full mixture, a binder gives a count of cells its specificity, and a DNA tag lets a protein measurement end as a count of DNA.

The table shows a pattern. Whenever a molecule lacks the two nucleic-acid tricks, methods try to borrow them, most often by attaching a DNA tag to a binder so that the measurement ends as a count of DNA. The proximity extension assay and CITE-seq in the table both rest on that one move.

+ Part II · The nucleic-acid layers

Reading by copying and pairing.

At the sequencing centers funded by the US National Human Genome Research Institute, the cost of producing a human genome sequence fell much faster once the centers moved to instruments that read millions of copied DNA fragments in parallel, and Chapter 5 gives the figures. The genome, the epigenome and the transcriptome can all be read on such instruments, because their molecules can be copied and recognized by pairing. This part shows how a genome is read in fragments and put back together against a reference, and why laboratories still run three generations of sequencer for three different questions. It then shows how chemistry and enzymes turn methylation into letters a sequencer can read, how an enzyme marks open chromatin for sequencing, and how counting the reads that land on each gene measures what a cell is using.

+ The questionHow are three billion letters read when no instrument reads a chromosome end to end?

Shatter, read, reassemble

Chromosome 1, the largest human chromosome, is a single DNA molecule about 250 million base pairs long. No sequencer reads anything close to that in one pass: the most common instruments read a few hundred bases at a time, and even the long-read machines described in Chapter 5 typically read tens of thousands. Every genome is therefore read the same way. The DNA is broken into pieces, an instrument reads the sequence of each piece, and software puts the pieces back in order. The sequence that an instrument reports for one piece is called a read.

The laboratory half of that process is called library preparation. A laboratory extracts DNA from cells, breaks it into fragments and fits each fragment with short synthetic sequences, called adapters, at both ends. The adapters are what the instrument grips and where the reading starts, so every fragment presents the same handle, whatever part of the genome it came from. The result is a library: hundreds of millions of fragments, each a random sample of the genome and each ready to be read.

The computational half turns the reads back into a genome. Most human sequencing uses alignment, in which software matches each read to the position it best fits in a reference genome, the agreed standard sequence that serves as a coordinate system. Where no suitable reference exists, software assembles the reads from scratch by finding the overlaps between them, the way a torn page can be reconstructed from overlapping scraps. Both approaches rely on the reads being long enough, and unique enough, to find their place.

Every base is read many times

A single read is not trusted, for two reasons. The first is error. Sequencers report a quality score for every base on a scale in which every 10 points is a tenfold step, so Q30 means an estimated one chance in a thousand that the base is wrong, and Q20 one chance in a hundred. Even at Q30, a genome read once would contain millions of errors. The second reason is sampling. Fragments land on the genome at random, so some positions are covered many times and some not at all, the same way raindrops leave some paving stones soaked and others dry.

The answer to both is coverage, the average number of reads covering each position. Genome laboratories also call it depth, a different sense of the word from that of Chapter 2, where depth is how scarce a molecule a method can still detect. A standard clinical or research genome is sequenced to about 30-fold coverage, written 30×. At that coverage, a position read by 28 reads, 14 showing one letter and 14 another, gives a confident call. A position that happens to be read by only 4 reads does not, which is why coverage is reported as a distribution and not just as an average.

What a variant looks like in the data

Once the reads are stacked against the reference, differences stand out, and their pattern says what they are. A heterozygous variant, present on only one of the two parental copies of a chromosome, appears in roughly half the reads at that position, and a homozygous variant, present on both copies, appears in nearly all of them. A random sequencing error usually appears in one read and nowhere else, because such errors rarely repeat at the same position across independent fragments. Systematic errors can recur, for example in runs of a single base, in duplicated fragments or in reads misplaced in repeats. The software that makes the calls therefore also weighs duplicate reads and strand balance, which is whether the supporting reads come from both strands of the DNA. It also weighs mapping quality, the confidence with which each read was placed.

That is the logic of variant calling: a variant is a difference that recurs in many independent reads at the same position, at a proportion consistent with one or two copies. The software combines those checks with the number of supporting reads and their base qualities, and it reports each call with a confidence of its own. The output of genome sequencing is therefore not a text read from start to finish but a list of differences from the reference, each with a measure of how sure the software is.

07 — Reading a genome in fragments

DNAReadoutFocal detail

The reference has blind spots

A reference genome is a coordinate system, and its gaps become blind spots. The reference used through most of the last two decades was missing about 8 % of the genome, mostly long repetitive regions that short reads could not assemble. In 2022 Nurk and colleagues of the Telomere-to-Telomere consortium published the first gapless human sequence, 3.055 billion base pairs long, which added roughly 200 million bases and nearly 2,000 predicted genes. The consortium assembled it from long reads of a single cell line rather than a person's cells, and its authors included staff of one maker of long-read instruments. The Y chromosome, which that sequence did not include, followed in 2023, when Rhie and colleagues of the consortium assembled it from a different person's DNA. In its statement of competing interests, the 2023 paper names one author as an employee of another maker of long-read instruments, and other authors as having received research support or travel funding from instrument makers.

A single reference also comes from very few people, so reads carrying DNA that the reference lacks either fail to align or align to the wrong place. The Human Pangenome Reference Consortium is building a reference from many people instead. In 2023 Liao and colleagues of the consortium published a first draft built from 47 individuals, and in May 2025 the consortium announced a second data release covering 232.

India has a project of its own. The Department of Biotechnology of the Government of India funded the GenomeIndia project. In January 2025 the government launched the project's data set and archived it in the Indian Biological Data Centre, India's national repository for biological data, through which researchers apply for access. In April 2025 Bhattacharyya and colleagues of the GenomeIndia Consortium described the data set in Nature Genetics as the whole genomes of 10,000 healthy, unrelated people from 83 Indian population groups. Their paper gives underrepresentation as the reason for the work, describing India as one of the highly underrepresented populations in global genomics.

Variant size decides what short reads can see

Variants come in sizes spanning more than eight orders of magnitude. A single-nucleotide variant changes one base, and small insertions and deletions change up to about 50. Structural variants, meaning deletions, duplications, inversions and insertions of more than 50 bases, can span thousands to millions of bases. Copy-number changes, and gains or losses of whole chromosomes, are larger still. A copy-number change alters how many copies of a long stretch of DNA a genome carries.

Short reads handle the smallest and the largest variants well. A single-base change sits comfortably inside a 150-base read, and a whole extra chromosome shows up as a proportional rise in coverage. The middle is hard. A structural variant several thousand bases long cannot be seen inside any one short read, so software has to infer it from indirect clues such as reads that align oddly or a change in coverage, and in repetitive regions the clues are often missing. Much of the reason long-read sequencing exists is to read across that middle range directly.

08 — Variant size against read length

DNAReadoutFocal detail

Genome, exome or panel

Not every question needs the whole genome. Less than 1.5 % of the genome encodes protein (Chapter 1). Sequencing just those regions, the exome, costs less and concentrates the reads where most known disease-causing variants sit. Targeted panels go further. A panel is a chosen set of tens to hundreds of genes read at very high coverage, often 500× to more than 1,000×, and that coverage is what lets a tumor panel find a mutation present in only a few percent of the cells in a biopsy. The trade is the usual one in omics, breadth against depth: a panel sees deeply into what it was designed for and nothing else, while a genome sees everything, less deeply.

+ The questionWhy do laboratories still run three kinds of sequencer?

One reading principle, three changes

A Sanger instrument reads up to about a thousand bases from each template, one template per capillary. A short-read instrument reads 100 to 300 bases from each of millions to billions of fragments at once, and a nanopore instrument can read a single DNA molecule more than 100,000 bases long. The three numbers describe three different capabilities, which is why laboratories still run all three kinds of sequencer: each kind answers a different question about the same DNA.

Every sequencer in routine use relies on the same fact: a base is identified by what it pairs with, or by the distinctive signal it produces as it passes a detector. Most machines copy the template strand one base at a time and detect each addition, while nanopore machines detect each group of bases as it moves through a pore. What changed across three generations of instruments is not that principle. It is how many molecules are read at once, whether they are copied before reading, and how long a stretch of one molecule can be read in one go. Each change answered a sharper question, and the questions are the clearest way to understand the instruments.

Generation one: what is this sequence?

Frederick Sanger and two colleagues at the Medical Research Council's Laboratory of Molecular Biology published the chain-termination method in 1977, demonstrating it on the DNA of a virus that infects bacteria. The method copies a single purified template in the presence of a small amount of modified bases that stop the copying wherever they are incorporated, so the result is a set of copies ending at every possible position. In 1977 the copies carried a radioactive label, and they were separated by length on a gel and read from X-ray film.

Later instruments replaced the radioactive label with a dye on each terminating base, one color for each of the four bases, and separated the fragments by length in a thin capillary, so the fragments pass a detector in order and the colors read out the sequence directly. Sanger sequencing reads up to about a thousand bases with very high accuracy, but only one template per capillary, so it answers one question well: what is the sequence of this particular piece of DNA? The Human Genome Project was built largely on it, at a cost of billions of dollars over more than a decade. Sanger sequencing is still the standard way to confirm a single variant found by other methods.

Generation two: what are all the sequences, and how many of each?

The step change came from parallelism. Short-read instruments attach millions to billions of library fragments to a glass surface, copy each fragment in place into a small cluster of identical molecules, and then read every cluster at the same time. Each chemical cycle adds one dye-labeled base to every growing strand and blocks any further addition. A camera then images the whole surface, the block is removed and the next cycle begins, so that after 100 to 300 cycles every cluster has produced a read.

The individual read is short and the chemistry has an error rate per base, but the output is enormous. The makers of current high-throughput instruments specify several terabases, or several trillion bases, per run, with most bases at Q30 or better. Parallelism did two things. It collapsed the cost, and it turned sequencing into a counting instrument, because when an instrument reads hundreds of millions of fragments, the number of reads that come from each gene, each allele or each organism is itself a measurement. RNA sequencing, single-cell methods and liquid biopsy all depend on that counting property more than on reading any one sequence.

Generation three: what is the whole molecule?

Single-molecule instruments skip the copying step. One approach watches a single polymerase at the bottom of a tiny well as it builds a partner strand, recording each incorporated base as a flash of light. Reading the same circular molecule many times over gives consensus reads. The maker of the instruments that work this way states that these reads are typically 15,000 to 20,000 bases long, with an accuracy of Q30 and above. The other approach pulls a single strand through a protein pore in a membrane and reads the changing electrical current as the bases pass. Its reads can exceed 100,000 bases, and in its own runs the maker of the nanopore instruments has reported reads of millions of bases. As of October 2026, that second maker reports a raw single-read accuracy of 99.75 %, or Q26, with the latest of its basecalling models, the software that turns the current into letters.

Reading a native molecule has a second benefit that is easy to miss. Because nothing is copied, chemical marks on the original DNA are still there to be detected, and both approaches report methylated cytosines directly from the signal, so a long read can carry sequence and epigenetic state together (Chapter 6). A long read can also show which variants sit together on the same parental copy of a chromosome, which geneticists call phasing. The question this generation answers is the one short reads struggle with: what does this whole stretch of molecule look like, in one piece, across repeats and structural variants, marks included?

The evolution at a glance

09 — Three generations, three questions

DNAModificationReadoutFocal detail

The cost of a genome, 2001 to 2022

The practical consequence of parallelism shows in the cost of a genome. The US National Human Genome Research Institute tracked the cost of producing a human genome sequence at the sequencing centers it funded, from about $95 million in 2001 to about $525 in 2022. Its figure is a production cost. It includes labor, reagents, amortized instruments and overheads, but it leaves out quality control, project management and any data analysis beyond the first processing of the raw data. The fall accelerated abruptly in 2008, when the centers switched from capillary instruments to massively parallel ones.

In announcements from 2022 to mid-2026, manufacturers gave figures of $80 to $345 per genome for their newest systems. The maker of the highest figure states that it is a list price for sequencing reagents only, quoted at 20× coverage rather than the usual 30×. The makers of the two lower figures do not say what their figures include or at what coverage they apply. None of the figures is comparable with the institute's production cost, but the direction of the fall is not in doubt. Chapter 23 gives the makers' figures as they stood in October 2026.

10 — Cost of sequencing one human genome

ReadoutFocal detail

+ The questionHow does a cell remember what it is when its sequence never changes?

Identity has to survive division

When a sequencer that copies DNA reads the genome of a liver cell and of a neuron from the same person, it reports the same letters for both. It also misses the chemical marks on the DNA that tell the two cells apart, because the enzyme that copies DNA before sequencing copies a marked base as an ordinary one. Yet when a liver cell divides, both daughters are liver cells. The sequence was copied, but since the same sequence also makes a neuron, something else must have been copied too: the record of which genes this cell keeps available and which it keeps silent.

That record is the epigenome, from the Greek for "on top of" the genome. It consists of chemical marks and physical packaging that sit on the DNA without changing its letters, and that a dividing cell passes to its daughters. Three kinds of record matter most for measurement. The first is DNA methylation, a small methyl group attached to cytosine. In most cell types it sits almost always where a C is followed by a G, a site written CpG, and the human genome has about 28 million such sites. Neurons are a notable exception, carrying substantial methylation at other sites too.

Methylation is copied through division by an enzyme that reads the mark on the old strand and adds it to the new one, which is what makes it a memory. The second kind of record is chromatin accessibility. DNA is wound around the protein spools called nucleosomes (Chapter 1), and a gene wound tightly is hard for the transcription machinery to reach. The third is histone modification: chemical marks that recruit or repel the machinery, carried on the histones, the proteins that make up each spool. Together the three decide which parts of the genome a cell can read.

Methylation: turning a mark into a letter

Because the methyl mark is lost at the first copy made before sequencing, the classic methods convert the question into one a sequencer can answer. Bisulfite treatment chemically converts every unmethylated cytosine into uracil, which is copied as T, while methylated cytosines are protected and stay C. After sequencing, each CpG in each read shows either C, meaning it was methylated, or T, meaning it was not.

Bisulfite chemistry is harsh and fragments much of the DNA, which matters when the starting material is scarce, as with the cell-free DNA in blood described in Chapter 14. Enzymatic conversion methods do the same job with enzymes instead. In a 2021 bench comparison, Vaisvila and colleagues tested one enzymatic method against bisulfite on genomic DNA, cell-free DNA and DNA from preserved tissue, across a range of input amounts. All 20 authors worked for the company that sells the enzymatic kit, which also funded the study. The authors reported that the enzymatic method damaged the DNA less than bisulfite did and worked from as little as 100 pg of DNA.

Newer approaches read methylation alongside the ordinary sequence in one library, and long-read instruments detect methylated bases directly from the native molecule without any conversion (Chapter 5). The conversion methods share one built-in limit. Standard bisulfite and enzymatic conversion both protect a related mark, 5-hydroxymethylcytosine, as well as 5-methylcytosine, the ordinary methylated C, so they cannot tell the two marks apart without an extra chemical step.

11 — Turning a chemical mark into a letter

DNAModificationFocal detail

The output of any of these methods is a methylation level: at each CpG, the fraction of reads that carried the mark. That fraction is a sampling estimate, and its precision depends on how many reads covered the site, because a fraction taken from 10 reads can move a long way by chance while a fraction taken from hundreds of reads moves much less.

Accessibility: an enzyme does the measuring

Accessibility is measured by asking an enzyme to find it. In 2013 Buenrostro and colleagues at Stanford University introduced ATAC-seq. The method loads an engineered transposase, an enzyme that inserts DNA, with sequencing adapters. Added to cell nuclei, the transposase can reach only DNA that is not wrapped around nucleosomes or covered by other proteins, and wherever it can reach, it cuts the DNA and inserts adapters in the same step. When the resulting fragments are sequenced and stacked on the genome, they form peaks where the chromatin was open: at active promoters, where a gene's transcription starts, at enhancers, which raise a gene's activity from a distance, and at other regulatory switches.

ATAC-seq is a clear case of the measuring-by-proxy principle of Chapter 3. Nobody observes chromatin opening. What is counted is where an enzyme was able to act, and the inference is that those regions were available to the cell's own machinery too.

12 — Reading which DNA is open

DNAReadoutFocal detail

Marks on the spools, and the fold

Histone modifications are read with antibodies. In chromatin immunoprecipitation followed by sequencing, or ChIP-seq, the laboratory cross-links the chromatin, chemically fixing proteins to the DNA they touch, then fragments it, pulls it down with an antibody against a particular mark and sequences the recovered DNA. In 2019 Kaya-Okur and colleagues at the Fred Hutchinson Cancer Research Center published CUT&Tag, which instead uses the antibody to guide the adapter-inserting enzyme to the mark. They profiled one histone mark from as few as 60 cells, far fewer than ChIP-seq needs. Both methods inherit the strengths and weaknesses of binding: they find exactly the mark the antibody recognizes, and they are only as specific as the antibody.

The genome also has a three-dimensional shape, and regulatory switches often act on genes far away along the sequence but close in space. Chromosome conformation methods measure that shape. Lieberman-Aiden and colleagues at the Broad Institute and the University of Massachusetts Medical School introduced Hi-C in 2009. In Hi-C the laboratory cross-links DNA that is touching in the nucleus, cuts it, joins the cut ends that the cross-links hold together and sequences the junctions. The frequency of each junction shows how often two distant regions sit together.

Why methylation reaches beyond the cell

Methylation patterns differ between tissues and change in disease, and they survive in DNA fragments after cells die, which makes them useful far from the cell they came from. Each fragment of DNA in the blood carries the methylation of the tissue that released it. The methylation of many fragments read together is the basis of methods that estimate which tissues cell-free DNA came from, and of blood tests that look for the methylation signatures of cancer, patterns across many sites that mark the disease (Chapter 14).

Methylation at certain sites also changes steadily with age, and statistical models built on those sites, often called epigenetic clocks, estimate age from a blood sample. In 2013 Horvath, at the University of California, Los Angeles, published a clock that combines 353 sites and works across most tissues and cell types. He built and tested it on about 8,000 samples from 82 public methylation data sets, and in the test data its estimates had a median error of 3.6 years. As a competing interest, the paper declares a provisional patent application owned by the University of California, with Horvath as inventor. A clock gives an estimate, not a direct measurement of biological age.

+ The questionKnowing a gene is present, how do we know the cell is using it, and how much?

A measurement that is really a count

A messenger RNA molecule is transcribed from one gene's sequence when the cell uses that gene, and it ends in a tail of A bases. The more intensely the cell uses a gene, the more copies of its messenger RNA it holds at any moment, and biologists call that level of use the gene's expression. Measuring gene activity therefore means measuring how many copies of each RNA are present, a steady-state level that reflects both how fast each RNA is made and how fast it is degraded. A typical mammalian cell holds on the order of a few hundred thousand messenger RNA molecules, spread across roughly ten thousand genes that are active in that cell type, with some genes represented by thousands of copies and others by one or two.

RNA sequencing, or RNA-seq, measures those levels by counting. A laboratory extracts RNA from a sample, and most of it is ribosomal RNA, the structural RNA of the protein-making machinery, which carries no information about gene activity. The laboratory therefore removes the ribosomal RNA, or it captures the messenger RNA selectively by the string of A bases on its tail. Reverse transcriptase, an enzyme that copies RNA into DNA, turns the remaining RNA into complementary DNA, or cDNA, which is fragmented, fitted with adapters and sequenced exactly as a genome library would be (Chapter 4). Software then aligns the reads to the genome or to a catalog of known transcripts and tallies the number of reads landing on each gene.

13 — From tissue to a count table

DNARNAReadoutFocal detail

The result is a count table: genes down one side, samples across the top, and in each cell of the table the number of reads. That number counts sequenced fragments, not molecules. The fragment count is proportional to the number of molecules of the transcript, multiplied by the transcript's length and by how efficiently the transcript survived capture and copying. All the transcripts in a sample also compete for the same fixed number of reads, so that total is divided among them and each count is a share of it.

Why raw counts mislead

Two of those factors distort comparisons unless they are corrected. The first is sequencing depth, the total number of reads a sample receives. Genome laboratories use the word depth for the coverage of each position (Chapter 4), but in RNA-seq the word means the number of reads in a whole sample. If one sample receives 40 million reads and another 20 million, every gene in the first will have roughly twice the count for reasons unrelated to biology. The second is transcript length. A longer transcript is cut into more fragments, so at the same number of molecules it collects more reads.

14 — Raw counts and counts per kilobase

DNARNAReadoutFocal detail

The third factor, composition, is subtler. Because the total number of reads is fixed, a sample in which one gene becomes enormously abundant leaves fewer reads for everything else, and every other gene then appears to fall although nothing changed for it. Differential expression is the comparison of each gene's level between conditions, and the normalization methods used for it estimate a scaling factor from the genes that did not change, precisely to avoid that false fall. Chapter 13 meets the same problem in a different guise.

Differential expression ends in a list of the genes whose expression changed, and any share quoted from that list takes the list as its denominator. The 7 % of the opening is a share of this kind. In the statement of March 2018 that the opening quotes, NASA gave it as a share of the gene expression that had changed during spaceflight, so the figure counted a share of the genes whose expression had changed, not a share of the genome.

How many reads is enough

Depth decides which genes can be measured at all. A replicate is one of several independent samples of the same condition. The ENCODE consortium's data standard for bulk RNA-seq asks for about 30 million aligned reads per replicate. The consortium last updated the standard in June 2017, and as of October 2026 that version is still current. At 30 million reads abundant transcripts are counted precisely and rare ones are sampled sparsely: a gene that yields 5 reads in one sample and 10 in another has not necessarily doubled, because counts that small fluctuate by chance by about the square root of their value. Detecting modest changes in genes with few RNA copies needs more reads, more biological replicates or both. Replicates matter more than depth in most designs, because the variation between individuals or cultures is usually larger than the counting noise.

One gene, many transcripts

Most human genes can be spliced in more than one way. A gene's exons are the segments of it that are kept in a messenger RNA, and the cell can join different subsets of them into different messenger RNAs, which may encode different proteins. Short reads see only fragments of each transcript, so software has to infer the relative amounts of each version statistically from the reads that happen to span a particular junction. Long-read sequencing can read a whole transcript in one pass, from start to tail. In 2023 Al'Khafaji and colleagues at the Broad Institute published a library method that joins several cDNA molecules end to end before long-read sequencing. On the long-read sequencer they used, the method raised the output more than 15-fold, to nearly 40 million cDNA reads per run. The authors disclosed a licensed patent application on the method, and after the paper was submitted the Broad Institute entered a collaboration agreement on the research with the sequencer's maker.

Before sequencing: arrays and targeted assays

Before RNA-seq, laboratories measured gene expression with microarrays: glass slides carrying thousands of fixed DNA probes, each designed to pair with one gene's RNA. Labeled cDNA from the sample pairs with its probes, and the brightness of each spot shows how much of that RNA was present. Arrays exploit pairing, just as sequencing does, but they read intensity rather than counting, and intensity has a floor set by background and a ceiling set by saturation. Arrays also measure only what their probes were designed for. RNA-seq counts fragments instead of reading an intensity, and counting extends the range. RNA-seq also reads sequence, which finds transcripts nobody knew to look for.

Targeted assays remain important, especially in the clinic. Quantitative PCR measures a handful of chosen RNAs with high precision, and several established multigene tests that guide cancer treatment decisions are built on it. The general pattern recurs across omics: discovery uses the broad method, and a decision that will be made repeatedly uses a narrow, validated one.

What RNA can and cannot show

A transcriptome is a snapshot of the last few hours of the cell's decisions, averaged over every cell in the sample, and both limits matter. A tissue is a mixture, so a change in the bulk signal can mean that every cell changed or that the mixture of cell types changed, and Chapter 12 describes how single-cell methods separate the two. RNA is also not protein: the relationship between the two is imperfect, for reasons Chapter 15 sets out.

+ Part III · Beyond nucleic acids

Measuring what cannot be copied.

Glucose, galactose and fructose are built from the same atoms and have the same exact mass, 180.0634 daltons, so a mass spectrometer can weigh any of the three sugars to within about a thousandth of a dalton and still not say which one it has weighed. Proteins and metabolites are where the cell's decisions become action, and neither can be copied or read by pairing, the two properties that make nucleic acids easy to measure (Chapter 2). This part follows the two routes around that limit, weighing molecules and binding them, and ends with the small molecules, where saying which molecule is present is harder than detecting that something is there.

+ The questionIf proteins cannot be copied or paired, how does a mass spectrometer identify thousands of them in one run?

The layer that acts

The human genome has about 20,000 genes that encode proteins, yet the proteome is larger and more varied than that count suggests. The same gene can give rise to several different protein molecules, because RNA can be spliced in different ways and because finished proteins are cut, folded and decorated with chemical groups such as phosphates and sugars. Each distinct molecular form is called a proteoform. Proteins do most of the work that makes a cell what it is: they catalyze its chemistry, carry its signals, build its structure and defend it. Many of the differences that matter for that work, such as whether an enzyme has been switched on by a phosphate, exist only at this layer.

Proteomics has made steady progress on coverage. Each year the Human Proteome Project, an initiative of the international Human Proteome Organization, reviews the public mass-spectrometry and antibody evidence that each predicted human protein exists. In the project's 2025 report, Deutsch and colleagues found credible protein-level evidence for 93.6 % of the 19,435 proteins predicted from the genome. The open question is therefore no longer whether a protein can be detected somewhere, but how many proteins can be measured in one sample, how precisely and how cheaply.

Weighing peptides instead of proteins

A mass spectrometer measures the ratio of a molecule's mass to its electric charge, the m/z of Chapter 3, so the molecule first has to become a charged particle in a vacuum. For proteins and peptides this is almost always done by electrospray. The liquid emerging from a chromatography column is sprayed from a fine needle held at high voltage, the charged droplets shrink as the solvent evaporates, and the dissolved molecules are left carrying one or more extra protons.

Electric or magnetic fields then sort the ions, and different instruments do it differently. A quadrupole passes only ions of a selected m/z through an oscillating field, so it works like a tunable filter. A time-of-flight analyzer gives every ion the same push and times its arrival, because lighter ions fly faster. An orbital trap lets ions circle a central electrode and measures how fast they oscillate, which depends on their m/z. One maker specifies the mass accuracy of its high-resolution instrument as better than 3 parts per million (Chapter 3). A few parts per million is enough to distinguish molecules whose masses differ in the third decimal place.

Most proteomics is done bottom-up: the proteins are first digested with trypsin into short chains of amino acids called peptides. Trypsin is an enzyme that cuts after the amino acids lysine and arginine, written K and R in the one-letter code. Peptides suit the instrument better than whole proteins do. Whole proteins are large, carry many charges and break unpredictably, whereas peptides of roughly 7 to 30 amino acids ionize well, separate well by chromatography and break in predictable places, so their sequences can be read from their fragments.

Liquid chromatography separates the peptides over a gradient lasting tens of minutes to hours, so that at any moment only a fraction of them enter the instrument. For each slice of time the instrument records a first spectrum, MS1, of all the peptide ions present. It then selects ions, breaks them by collision with gas and records the masses of the fragments in a second spectrum, MS2, and a single run records tens of thousands of MS2 spectra.

15 — Bottom-up proteomics

ProteinReadoutFocal detail

Reading a peptide from its fragments

A peptide breaks mostly along its backbone, between one amino acid and the next. Each break produces two pieces, one containing the start of the peptide and one containing the end, and across the many copies of a peptide in the instrument the breaks occur at every position. The pieces containing the end, called y ions, form a ladder: y1 is the last amino acid, y2 the last two, and so on. The difference in mass between consecutive rungs is the mass of one amino acid, so the gaps, read in order, give the sequence from its end.

16 — Reading a peptide from its fragments

ProteinReadoutFocal detail

From spectra to a protein list

In practice laboratories do not read the ladder by eye. Search software takes every protein sequence in a database, cuts it into peptides by computation, predicts the fragment spectrum of every resulting peptide and scores each observed spectrum against the candidates. It reports the best-scoring match if that score clears a threshold, and the threshold is set statistically. The software searches the real database and also a decoy database of reversed or scrambled sequences that cannot be present in the sample, so the number of decoy matches above a score estimates how many of the real matches above it are false. Laboratories typically report results at a 1 % false discovery rate, the expected share of false matches among those accepted, for peptides and again for proteins.

A further step, protein inference, turns peptides into proteins, and it is the least certain step in the chain. Many peptides are shared between related proteins or between proteoforms of one gene, so the peptides found often leave open which member of a family was present. That ambiguity is why results list protein groups, each the set of proteins that the same peptides could have come from, rather than single proteins.

Choosing what to fragment

The instrument cannot fragment every ion at once, so it has to choose. In data-dependent acquisition it selects the most intense ions in each MS1 scan, and that choice is partly random, so a peptide measured in one run may be skipped in the next, which leaves gaps in the data. In data-independent acquisition it fragments everything within successive m/z windows and relies on software to untangle the mixed fragment spectra, so the result is more complete and more reproducible across samples.

A 2024 method study shows what narrow windows on a current instrument can reach. Guzman and colleagues at the University of Copenhagen ran narrow-window data-independent acquisition on one instrument model and measured about 10,000 human protein groups from a digest of one human cell line in a 30-minute run. Fourteen of the study's 22 authors were staff of the company that makes the instrument.

Plasma is harder, for the dynamic-range reason Chapter 2 set out: albumin and a few dozen other abundant proteins can take almost all of the instrument's capacity. Laboratories therefore remove the most abundant proteins before measurement, or enrich lower-abundance proteins on the surfaces of nanoparticles. Developers of the newest enrichment workflows report several thousand proteins.

The size of the gain from enrichment shows in a 2025 comparison of six plasma workflows by Beimers and colleagues at the University of Wisconsin–Madison, who ran each workflow five times on one commercial plasma sample. Plasma measured directly, with no depletion or enrichment, gave about 600 proteins per run, and plasma enriched on the nanoparticles of one company, Seer, gave about 4,500. A trade report of the study states that Seer initiated the comparison and paid for some of its assays and for the plasma, and that at the time the senior author sat on Seer's scientific advisory board and consulted for Thermo Fisher Scientific. Völlmy and colleagues at Roche compared common plasma workflows in a study the company funded, and in 2026 they reported that enrichment identified up to 4,600 proteins. The next chapter describes the other route to the faint end.

How much, and in what form

Mass spectrometry measures ion signal, and different peptides ionize with different efficiency, so signal is not directly a quantity. Relative quantity compares the same peptide across samples, and it comes from signal intensity or from labeling each sample with an isotope tag and measuring the samples together. Some tags differ in mass and are told apart in the first spectrum, whereas isobaric tags weigh the same and are told apart by reporter ions released on fragmentation. Absolute quantity requires an internal standard, a known amount of an isotope-labeled copy of the peptide added to the sample. Clinical mass spectrometry assays are built this way, and each one is narrow, targeted and calibrated.

Modifications can be studied directly, because a phosphate or a sugar adds a known mass to the peptide that carries it. Phosphoproteomics enriches phosphorylated peptides before measurement to map which proteins in signaling pathways carry a phosphate. Top-down proteomics weighs and fragments intact proteins, so it measures proteoforms whole, at lower throughput.

+ The questionWhere mass spectrometry cannot see the rare proteins in blood, how do affinity methods reach them, and what do they give up?

The sandwich immunoassay and the limit on mixing

A signaling protein such as interleukin-6 circulates in a healthy person's blood at a few picograms per milliliter, at the faint end of the range of plasma concentrations that Chapter 2 described and below the reach of the mass spectrometry of Chapter 8. A long-established way to measure a single protein at such low concentrations is the sandwich immunoassay, often called an ELISA, for enzyme-linked immunosorbent assay. One antibody fixed to a surface captures the protein, and a second antibody, which recognizes a different part of it, binds on top. The second antibody carries an enzyme that converts a colorless substrate into a colored or light-emitting product.

Two features make the method work. Requiring two antibodies to bind the same molecule makes the signal specific, because a molecule that happens to stick to one antibody rarely also binds the other. The enzyme then amplifies the signal, because each bound enzyme turns over thousands of substrate molecules, and routine assays reach picograms per milliliter this way.

The obvious next step is multiplexing, measuring many proteins in one well, and mixing many antibody pairs for it runs into a combinatorial problem. With 10 pairs in a mixture, each detection antibody can meet any of the 10 capture antibodies, which makes 100 combinations. With 100 pairs there are 10,000 possible combinations, and every one that forms by accident produces false signal. Such accidental binding is called cross-reactivity, and it limits how many antibody pairs a conventional multiplexed immunoassay can mix on its beads or spots. In a 2014 development study of a 96-protein panel, Assarsson and colleagues put that limit at around ten proteins per well, and most of the authors were employees of the company commercializing that panel. The panel itself used the method of the next section.

Turning a protein into DNA

The methods that got past that limit did it by converting the protein measurement into a DNA measurement, borrowing the copying and pairing that proteins lack (Chapter 2). In a proximity extension assay, two antibodies target each protein, and each antibody carries a short single strand of DNA. When both antibodies bind the same protein molecule, their DNA strands are held close enough to pair with each other at their tips. A polymerase then extends them into a new double-stranded DNA barcode unique to that protein, and the barcodes are copied and counted, by quantitative PCR or by sequencing.

17 — Turning a protein into DNA

DNAProteinReadoutFocal detail

The design solves the mixing problem. An antibody that sticks where it should not does no harm unless its partner also lands on the same molecule, close enough for the DNA tips to meet. Because the readout is DNA, thousands of assays can also share one tube and be read together on a sequencer. As of October 2026 the maker of the proximity extension assay specifies that its current panel measures about 5,400 proteins from 2 µL of plasma.

An approach related to the proximity extension assay adds purification steps to suppress background. Feng and colleagues described it in a 2023 development study of a panel of about 200 proteins, and several of the authors were employees and shareholders of the company that makes it. In their own comparison they reported detection limits in the attomolar range, several orders of magnitude below conventional immunoassays. At one attomolar, a milliliter holds about 600 molecules of the target.

Aptamers: binders made of nucleic acid

A second large-scale route uses aptamers instead of antibodies. Chapter 3 introduced aptamers as short folded nucleic acids that bind by shape. The aptamers used for proteomics are short single strands of chemically modified DNA, each folded into a shape that binds a particular protein. Developers find each aptamer by selection: they expose a vast random library of strands to the target, keep and copy the strands that bind, and repeat the cycle until strong binders dominate. Because the binder is itself a nucleic acid, it can be read directly by hybridization or sequencing once the unbound strands are washed away. In 2010 Gold and colleagues described an early version of one company's aptamer platform, which measured 813 proteins and read the aptamers by hybridization to a DNA microarray. Most of the authors were employees of the company that developed it.

The same company stated at the launch of a new aptamer panel in 2023 that the panel gives about 11,000 protein measurements from 55 µL of plasma, a figure that counts measurements rather than distinct proteins. An aptamer platform typically uses one binder per protein rather than a pair, so each of its measurements depends more heavily on the specificity of a single binder than a measurement that needs two binders to agree.

What each route sees

Mass spectrometry and affinity methods answer different questions, and the difference lies in how each one works rather than in how mature it is. Mass spectrometry identifies proteins by their sequence, needs no prior list, and sees modifications and variants, but it is shallow in plasma because abundant proteins consume its capacity (Chapter 8). Affinity methods reach deep into the low-abundance range, and the maker of the proximity extension assay states that its platform runs thousands of samples a week. Yet affinity methods measure only the proteins they were designed against. An affinity method's identity claim is only as good as the evidence that each reagent binds its intended target and nothing else, because the method reports whatever the reagent binds, which can include complexes, fragments or a related protein.

18 — Where each proteomics route can see

ProteinReadoutFocal detail

A subtle consequence matters for genetics. A genetic variant that changes the part of a protein an antibody or aptamer recognizes can change how well the reagent binds without changing how much protein is present. In a large study, that change in binding shows up as a strong genetic effect on the measured level, and it can be mistaken for real regulation of the protein.

Proteomics at population scale

Affinity methods are what made population-scale proteomics possible, and UK Biobank shows the scale. It holds samples from about 500,000 participants, and a consortium of 13 biopharmaceutical companies funded its Pharma Proteomics Project, which measured 2,923 proteins by proximity extension in plasma from 54,219 participants. Sun and colleagues analyzed those data in 2023 and mapped more than 14,000 associations between genetic variants and protein levels. Most of the 56 named authors gave a company as their affiliation.

In January 2025 UK Biobank announced an expansion that aims to measure up to 5,400 proteins in 600,000 samples on the maker's current proximity extension panel. A consortium of 14 biopharmaceutical companies funds the expansion. The samples are one from each of the roughly 500,000 participants and 100,000 second samples taken from the same people up to 15 years later. UK Biobank plans to release the data in stages from 2026, with the full dataset expected by 2027, and the study was in progress as of October 2026. Studies like these link genetic risk to the proteins that carry it, and they are a major source of new drug targets.

The third route: reading proteins one molecule at a time

Chapter 2 noted that single-molecule detection is the one route that could remove the advantage nucleic acids enjoy, and several groups are pursuing it for proteins. One company has shipped a protein-sequencing instrument since 2023 and launched an upgraded version in January 2025. The instrument holds peptides on a chip and detects fluorescent recognizer proteins as they bind the amino acid at each peptide's exposed end, while enzymes trim the peptide one residue at a time. Reed and colleagues described the method in 2022 in a technology demonstration on synthetic peptides and protein digests, and most of the authors were the company's own scientists.

The company's newer results come from its next instrument, which had not launched as of October 2026, and the company reported them in its own announcements. In September 2026 the company reported detecting 18 of the 20 amino acids with a developmental kit run on that instrument, and in October 2026 it reported detecting 19 in the instrument's development program. The company expects to show all 20 by the end of 2026 and to launch the instrument commercially in the second quarter of 2027.

Academic groups are pursuing a different route through nanopores, the pores that some DNA sequencers read through (Chapter 5). In a 2024 bench study, Motone and colleagues at the University of Washington used a molecular motor to pull synthetic protein strands hundreds of amino acids long through a nanopore, and the electrical signal was sensitive to single amino acids. The team read the strands on flow cells made by Oxford Nanopore, and the paper's bibliographic record lists Oxford Nanopore among the study's funders. Rereading each molecule ten times raised the team's accuracy from 28 % to 61 % at telling apart strands that differed at one position, in a test with 20 possible variants where chance alone would give 5 %. The study classified known differences rather than reading an unknown sequence.

A second company takes another approach. In January 2026 it opened early access to its platform as a fee-for-service offering, starting with a tau assay that measures the protein's proteoforms, the distinct molecular forms that Chapter 8 defined. It says the platform is designed to map up to 10 billion intact protein molecules by probing each one repeatedly. The first company and the academic groups have published demonstrations, whereas the first company's newer results and the second company's design figure are the companies' own announcements. As of October 2026, none of these routes measures a plasma proteome at a depth or cost that competes with mass spectrometry or affinity methods. Chapter 24 separates what each of these routes has shown from what its developers have announced.

+ The questionWhy is the layer closest to the phenotype also the hardest to identify?

Small molecules, enormous variety

The Human Metabolome Database listed 217,920 metabolites in its 2022 release, yet most of its entries are molecules expected or predicted to occur in people rather than molecules detected in human samples. Metabolites are the small molecules of life, typically under about 1,500 daltons: sugars, amino acids, fatty acids and other lipids, nucleotides, vitamins, hormones, and the intermediates of every pathway that makes or breaks them. A blood or urine sample also carries molecules the body did not make, such as drug residues, food components, pollutants and the products of gut microbes.

Two features set this layer apart. Metabolites are the direct substrates and products of enzymes, so they reflect what the cell's machinery is actually doing rather than what it could do. They are the closest molecular layer to the phenotype, the observable outcome of a cell or an organism, and a change in diet, drug or disease shows up here within minutes. Metabolites also have no template: nothing in the genome encodes a metabolite the way it encodes a protein's sequence, so the list of what should be present cannot be fully predicted from the genome. Every identification has to rest on evidence from the chemistry of the molecule itself.

Three instruments, three trade-offs

Liquid chromatography coupled to mass spectrometry, LC-MS, covers the widest range of metabolites, because different columns retain polar or fatty molecules and running the ion source in positive and negative mode captures molecules that prefer to gain or to lose a proton. Gas chromatography with mass spectrometry, GC-MS, handles molecules that are volatile or can be made volatile by chemical derivatization, a reaction that attaches groups which make them evaporate more easily. Its fragmentation is so reproducible that large reference libraries of spectra exist for matching.

Nuclear magnetic resonance, or NMR, works on a different principle entirely. It is the main exception to the five physical properties of Chapter 3, and its spectrum reports the atomic environments in a molecule. NMR is less sensitive than mass spectrometry, typically reaching micromolar rather than nanomolar concentrations. It is quantitative without standards for each compound, it does not destroy the sample and it is highly reproducible between laboratories, so it suits very large studies of the most abundant metabolites and lipoproteins.

Identifying a metabolite takes more than its mass

Chapter 3 noted that mass alone rarely identifies a molecule, and metabolomics is where that limit matters most. Glucose, galactose and fructose, the three sugars of this part's opening, all have the formula C₆H₁₂O₆ and the same exact mass, 180.0634 daltons. A mass spectrometer that weighs them to within about a thousandth of a dalton still cannot tell them apart, because they are isomers: the same atoms, arranged differently. The same problem appears across the whole metabolome, since many lipids differ only in where a double bond sits and many drug metabolites differ only in which position carries a hydroxyl group.

19 — One mass, several molecules

MetaboliteReadoutFocal detail

Identification therefore needs a second property. Chromatography separates many isomers, so a molecule's retention time, the moment it emerges from the column, helps to identify it, and so does the way it fragments. The decisive evidence comes from running an authentic standard, a pure sample of the suspected compound, under the same conditions and showing that it matches on at least two independent properties.

An untargeted LC-MS run records whatever ionizes rather than a chosen list, and it detects features: signals each defined by an m/z and a retention time. A typical dataset contains tens of thousands of them. Most are not distinct compounds. The same molecule appears as several features because it picks up different ions such as sodium or potassium, because of natural carbon-13 isotopes, and because some molecules break apart in the ion source, so a count of features overstates the number of compounds present.

Chi and colleagues at the Jackson Laboratory for Genomic Medicine set out to separate real compounds from these duplicate features. They reanalyzed 61 public LC-MS datasets from blood, using the patterns that isotopes and added ions make to group the features that come from one compound. They estimated that each dataset detects about 1,000 to 2,000 compounds with high confidence, that more than half of these compounds are unknown, and that fewer than half match an entry in the Human Metabolome Database. The analysis was posted as a preprint in 2025 and published in a journal in 2026.

In 2007 Sumner and colleagues proposed confidence levels for reporting identifications on behalf of a working group of the Metabolomics Standards Initiative, and those levels remain the working vocabulary:

20 — From feature to identified metabolite

MetaboliteFocal detail

Targeted metabolomics: the version already in routine use

The opposite approach, targeted metabolomics, measures a predefined set of metabolites, each with an isotope-labeled internal standard of the kind Chapter 8 described, and reports absolute concentrations. It is less exploratory and far more reliable, and it is the form in which metabolomics entered routine medicine long before the word existed. Newborn screening measures amino acids and acylcarnitines, the forms in which carnitine carries fatty acids, in a dried blood spot by tandem mass spectrometry, the two-stage measurement of Chapter 8, to detect inherited metabolic disorders. Laboratories developed the method in the early 1990s, and public screening programs adopted it from the late 1990s. A screening panel is the list of conditions a program tests every newborn for. As of October 2026 the method covers a large share of the conditions on panels such as the US recommended panel.

Handling is part of the measurement

The metabolome changes within seconds to minutes, and it keeps changing after the sample is taken. Blood cells in an unprocessed tube continue to consume glucose and release lactate, red cells that break release their contents, and repeated freezing and thawing degrades unstable molecules. Biological context adds further variation, because fasting state, time of day, recent meals, exercise and medication all move metabolite levels. A metabolomics study that does not standardize collection, processing time, temperature and storage therefore measures its own sample handling as much as the biology of its subjects.

Lipids and glycans

Two sub-fields deserve separate mention because they matter in both biology and manufacturing. Lipidomics measures the thousands of lipid species in membranes, stores and signaling pathways; it uses mass spectrometry and faces an extreme version of the isomer problem. Glycomics measures glycans, the branched sugar chains attached to proteins and lipids. No template encodes glycans: enzymes build them, which makes them heterogeneous, and their branching produces isomers that share a mass and often emerge from the column together. A standard workflow releases glycans from proteins with enzymes before chromatography and mass spectrometry. In therapeutic antibodies the glycan pattern affects how the drug behaves in the body, which is why manufacturers monitor it during production as a critical quality attribute, a property that should stay within set limits to ensure the product's quality.

+ Part IV · Cells, space and shed material

From averages to individuals.

An oil droplet that holds one cell and one gel bead is where a widely used form of single-cell sequencing labels the cell's molecules. The bead carries a DNA barcode found on no other bead, and that barcode marks every RNA molecule captured from the cell, so that millions of cells can be pooled, sequenced together and sorted back into individuals afterwards (Chapter 12). Everything before this part measured a sample as a whole. This part measures the sample's parts: individual cells, where they sit in a tissue, the other organisms living alongside them, and the vesicles and DNA fragments that cells release into the blood. The recurring problem changes from identification to detection limits.

+ The questionIf a tissue is a mixture, how do we measure each cell instead of the average?

The averaging problem

A blood sample holds billions of cells of dozens of types. When a laboratory grinds up such a sample and measures one protein, the result is a single number: the average across every cell, weighted by how common each type is. That average hides the distinction that matters most in cell biology, between a change inside cells and a change in which cells are present.

Suppose the protein is a marker, one whose level distinguishes a cell type or state, and the bulk measurement shows it rising by 10 %. Either every cell now carries 10 % more of it, or 10 % more of the cells are of a type that carries it, or a rare population has expanded dramatically while everything else stayed still. In the bulk measurement the number is the same in all three cases, because grinding the cells together erases which cell carried which molecules. Only a measurement of each cell can tell the three apart.

Cytomics is the measurement of many properties of individual cells across very many cells. Its oldest and most widely used instrument is the flow cytometer. The flow cytometer established three ideas, labeling cells with specific binders, reading them one at a time and counting them, and all three return in the single-cell sequencing of the next chapter and in the vesicle measurements of Chapter 14.

Inside a flow cytometer

A flow cytometer has three subsystems: fluidics, optics and electronics. The fluidics inject the cell suspension as a thin core stream inside a faster-moving sheath of fluid. The sheath accelerates and narrows the core, a process called hydrodynamic focusing, so the cells are forced into single file and pass the measurement point one at a time.

The optics shine one or more lasers across the stream, and each passing cell scatters some of the light. Light scattered at small forward angles grows with the cell's size and is called forward scatter. Light scattered to the side grows with the cell's internal complexity, such as its granules and the shape of its nucleus, and is called side scatter. Cells that carry fluorescent labels also emit light at longer wavelengths, which a series of mirrors and filters routes to separate detectors.

The electronics convert each flash into a pulse, a signal that rises as the cell enters the beam and falls as it leaves, and they digitize the pulse's height, area and width. Each pulse that rises above the instrument's threshold is recorded as an event, usually one cell, and each event is one row in a table that holds its forward scatter, its side scatter and one value per fluorescence detector. Conventional instruments record thousands to tens of thousands of cells per second. Sorting instruments add a final step: they break the stream into droplets and deflect the droplets that contain wanted cells into separate tubes.

21 — Inside a flow cytometer

ReadoutFocal detail

Labels, panels and controls

The fluorescence comes from antibodies against cell-surface or internal proteins, each linked to a fluorescent dye. A set of such antibodies chosen to be measured together is called a panel, as a chosen set of genes is in Chapter 4. Most of the skill in cytometry lies in panel design. Designers pair bright dyes with proteins present at few copies per cell, so that those proteins still rise above background, and dim dyes with abundant ones. Markers found on the same cells are given the dyes whose emissions overlap least, for a reason the next section explains. Every panel also needs controls, the samples and stains that show where background ends: unstained cells, cells stained with every antibody but one, and a dye that marks dead cells, because dead cells bind antibodies nonspecifically.

More colors, more overlap

Each dye emits across a broad band of wavelengths, so a detector meant for one dye also catches some light from its neighbors. Analysis software corrects this spillover mathematically, a step called compensation, but each correction adds noise, and the problem grows quickly as colors are added. With one filter per dye, conventional instruments become impractical at a few dozen colors.

Two designs reach beyond that limit. Spectral cytometers record each cell's full emission spectrum across many narrow detectors and then unmix it, using each dye's reference spectrum to work out how much light each dye contributed. In a 2024 report, Konecny and colleagues described a 50-color panel, developed on a spectral instrument, for profiling human immune cells. Two of the report's five authors were employees of BD Biosciences, which made the instrument. In 2026 another manufacturer stated that its new spectral instrument can run 60-color panels, and offered the instrument through an early-access program (Chapter 23).

Mass cytometry abandons light entirely. Its antibodies carry tags made of heavy-metal isotopes, and the instrument vaporizes each cell, ionizes its atoms and counts the metal tags in a time-of-flight mass spectrometer of the kind described in Chapter 8. Metal masses barely overlap and cells have no natural metal background, so many parameters can be read cleanly. In a 2022 review of mass cytometry, Iyer, Hamers and Pillai of the University of Miami report that, on average, about 60 parameters can be analyzed robustly at once. The costs are three: the instrument reads hundreds rather than thousands of cells per second, it destroys every cell it measures, and a substantial fraction of the cells is lost before measurement.

22 — Separating colors: three designs

ReadoutFocal detail

From events to populations

Analysts select populations by gating. A gate is a boundary drawn on a two-dimensional plot of one parameter against another, and a sequence of gates narrows the events step by step: for example living cells, then single cells, then T cells, then a subset of T cells. With dozens of parameters, manual gating gives way to algorithms that group cells with similar measurements into clusters, and to two-dimensional maps of the high-dimensional data. Either way the result is a set of populations, each reported as a fraction of a parent population or as an absolute count per microliter.

Rare populations bring a statistical limit that no instrument removes: counts of independent events fluctuate by about the square root of the count, which means that the precision of a population estimate depends on how many cells were counted in that population.

Why cytometry is an omics method

A cytometry run that measures 30 to 50 proteins on each of millions of cells describes the composition and state of a cell population in more detail than most bulk proteomics can. Methods that bridge to sequencing go further, because antibodies that carry DNA barcodes instead of dyes can be read by a sequencer together with each cell's RNA, as Chapter 12 describes. These methods descend directly from cytometry, because they keep its three steps: they label with a specific binder, read one object at a time, and count.

+ The questionCan we read the whole transcriptome of each cell, and still know where in the tissue it came from?

Labeling instead of separating

In November 2024 the Human Cell Atlas consortium published a collection of its studies in Nature, and its data portal then held data on about 62 million individual cells from about 9,100 donors. Data of that kind needed a different method from cytometry, because measuring the whole transcriptome of each cell is a problem of a different scale from measuring dozens of proteins per cell. A single cell holds only about 10 pg of RNA, and early methods handled one cell per tube, which limited studies to hundreds of cells. The methods that scaled to millions did so by separating cells only long enough to label them. If every RNA molecule from one cell carries the same DNA tag, all the cells can be pooled and sequenced together, and the reads can be sorted back into cells afterwards.

A widely used implementation uses droplets: a microfluidic chip combines a stream of cells, a stream of gel beads and oil, and produces thousands of droplets per second. Every bead carries millions of copies of a DNA strand with three working parts. The first is a cell barcode, which is identical on every strand of that bead and different from every other bead. The second is a unique molecular identifier, or UMI, a short random sequence that differs from strand to strand, and the third is a string of T bases that pairs with the A-rich tail of messenger RNA.

Cells are loaded dilute, so most droplets hold no cell and an occupied droplet ideally holds one cell and one bead. Inside the droplet the cell is broken open and the strands capture its RNA, and reverse transcription then writes the barcode and the UMI into each cDNA molecule. Finally the droplets are broken, and everything is pooled, amplified and sequenced.

23 — Barcoding every cell

DNARNAReadoutFocal detail

Two labels, two jobs

The cell barcode answers the question "which cell did this molecule come from?", and the UMI answers a subtler one: "is this read a new molecule, or a copy of one already counted?" Amplification makes many copies of each captured molecule, and some molecules are copied more than others, so a count of reads would over-count the molecules that amplification happened to copy most. A count of distinct UMIs counts the original molecules instead.

Only a fraction of each cell's RNA molecules is ever captured and converted, so the data are sparse: most genes read zero in any given cell, even when the cell expresses them at low levels. Single-cell data are therefore powerful for telling cell types and states apart, and weak for measuring small differences in a single low-abundance gene.

Scale, and what it has produced

Throughput has grown by orders of magnitude. In October 2024 the maker of one droplet system stated that the system processes up to 2.5 million cells per run, at a cost as low as one cent per cell. Other methods need no droplets at all: they barcode cells by repeatedly splitting and pooling them in plates, adding one barcode segment per round, and their makers state that one kit can label a million or more cells.

The data from such experiments are gathered in shared resources as well as in the Human Cell Atlas portal. The Chan Zuckerberg Initiative runs one such openly available resource, CELLxGENE Census, and its November 2025 release held about 125 million unique cells from five species, about 100 million of them human, drawn from more than 1,800 datasets. Single datasets have reached the same scale. In a 2025 preprint, the company Tahoe Therapeutics described one dataset that profiled the RNA of 100 million cells from 50 cancer cell lines, each line exposed to about 1,100 combinations of drug and dose. The cells were barcoded by the split-and-pool method in a service laboratory run by one of that method's makers, and 18 of the preprint's 32 authors worked for that maker.

From a matrix to cell types

The output is a matrix of cells by genes that holds UMI counts. Analysis software groups cells with similar expression profiles into clusters, as cytometry software groups cells by their proteins (Chapter 11). It then draws the clusters on two-dimensional maps and labels each cluster by the marker genes it expresses. Cells caught at different stages of a continuous process, such as differentiation, can be ordered into trajectories.

Two artifacts recur. A droplet sometimes captures two cells, and the resulting doublet looks like a hybrid cell type that does not exist. RNA released into the suspension by broken cells is captured in every droplet, so it adds a faint background that all the cells share. Both are handled computationally, and both need to be checked.

Adding proteins and chromatin

The barcoding logic extends to other layers. For surface proteins, antibodies that carry DNA barcodes instead of fluorescent dyes bind the cell surface before the cell enters the droplet, and their barcodes are then captured and sequenced alongside the RNA. Stoeckius and colleagues described this method, CITE-seq, in 2017: it gives surface protein levels and the transcriptome from the same cell, joining cytometry's view of the cell to sequencing's. Chromatin accessibility can be read cell by cell too, by running the adapter-inserting enzyme of Chapter 6 on nuclei before barcoding, and some kits read RNA and accessibility from the same nucleus.

Putting cells back in place

Dissociating a tissue into single cells throws away where each cell sat, and location matters: a cell's neighbors decide much of what it does, and a pathologist reads disease from the structure of a tissue. Spatial methods keep each measurement's position in the tissue, by two routes that are converging.

Capture arrays place a tissue section on a slide covered with barcoded spots whose positions are known. Each spot captures RNA from the tissue directly above it, and the spot's barcode records where that RNA came from. Capture arrays read the whole transcriptome, and their resolution is set by spot size. Their makers state that some current arrays use features 2 µm wide or smaller, which is smaller than a cell, whereas earlier arrays used spots 55 µm across that pooled several cells.

Imaging methods leave the RNA in place and identify each molecule under a microscope. Panels of probes hybridize to their target RNAs over several rounds, and each gene's pattern of on and off signals across the rounds forms a code that identifies it. The resolution is subcellular and each molecule is placed individually, but only the genes in the designed panel are read. Makers had grown these panels from hundreds of genes to 5,000 or 6,000 by 2024, and in April 2025 one maker described a whole-transcriptome imaging panel that covers more than 18,000 RNAs. Protein imaging works the same way with barcoded antibodies and reaches a hundred or more markers per section.

24 — Keeping the map: two approaches

DNARNAReadoutFocal detail

Every spatial method must also decide which molecules belong to which cell, a step called segmentation that usually draws cell boundaries from a nuclear or membrane stain. Segmentation errors assign molecules to the wrong neighbor, and in dense tissue they are a leading source of false cell-to-cell signals.

+ The questionWhen a sample holds a thousand species, how do we tell whose DNA is whose?

A second genome, carried everywhere

A man weighing 70 kg carries an estimated 38 trillion bacterial cells and about 30 trillion human ones, so his body holds roughly as many bacterial cells as human cells. Sender, Fuchs and Milo of the Weizmann Institute of Science reached those figures in a widely cited 2016 estimate for a reference man aged 20 to 30. They worked organ by organ from published volumes and cell densities, and they placed most of the bacteria in the colon.

Collectively those microbes carry far more genes than the human genome, and they do chemistry the body cannot do alone. They digest fibers, make vitamins and short-chain fatty acids, modify drugs, and train the immune system, and the metabolites of Chapter 10 are partly theirs. So the question "what is this person's molecular state?" includes a question about other organisms' genomes. Metagenomics reads those genomes from a mixed sample without growing each organism in culture, which most of them resist.

Marker-gene sequencing: who is there, cheaply

Every bacterium carries a gene for 16S ribosomal RNA, about 1,500 bases long. Parts of the gene are nearly identical across all bacteria, while nine other parts differ between lineages and are called the hypervariable regions. That structure is ideal for a census: primers that pair with the conserved stretches amplify a hypervariable region from every bacterium in the sample at once, so sequencing the products shows which lineages were present and roughly in what proportions. Fungi are surveyed the same way with a different marker region.

The method is cheap and sensitive, and it has three known limits. It usually resolves organisms only to genus, the level above species, because closely related species share the same marker sequence. Amplification favors some sequences over others. And bacteria carry between one and about fifteen copies of the 16S gene, depending on species, which means a species with many copies looks more abundant than it is unless the analysis corrects for the number of copies.

Shotgun metagenomics: who is there, and what they can do

Shotgun metagenomics skips the amplification of a marker and sequences all the DNA in the sample, as the genome sequencing of Chapter 4 does, but without a single reference genome. Software assigns the reads to species, and often to strains, by matching them against genome databases, and it catalogs the genes directly. The gene catalog shows what the community is equipped to do, including which antibiotic-resistance genes it carries.

The costs are money and interference from the host. Deep shotgun sequencing costs several times more per sample than a marker census, although shallow shotgun sequencing narrows the gap. In clinical samples most of the DNA is often human, so most reads are spent on the host unless the laboratory depletes human DNA first. Sequencing RNA instead of DNA, which is called metatranscriptomics, adds which genes the community is actively using, the same step from genome to transcriptome that Chapter 7 describes.

25 — Two ways to read a community

DNAFocal detail

A share can fall while the count stays the same

Both methods produce relative abundances: each organism's share of the reads. Because the shares must sum to 100 %, a rise in one organism lowers the shares of all the others, even when their numbers have not changed.

The effect is the composition effect that Chapter 7 met in RNA sequencing, where one abundant gene absorbs reads from all the others, and the remedies are similar. A laboratory can add a known quantity of a reference organism or DNA, called a spike-in, to each sample before extraction, so that shares can be converted into absolute amounts. It can measure the total microbial load independently, for example by quantitative PCR or by counting cells. Or its analysts can use statistical methods designed for proportions, which compare ratios between organisms rather than shares.

26 — Shares hide the total load

ReadoutFocal detail

Low biomass and contamination

Laboratory reagents are not sterile at the level of DNA: extraction kits and water carry traces of bacterial DNA. In a stool sample, which is rich in microbes, those traces vanish among the sample's own DNA. Blood, spinal fluid and tissue biopsies are low-biomass samples, with little microbial DNA of their own, and in them the traces can dominate the result. A few published claims of microbes in supposedly sterile sites have been traced to contamination. The defense is procedural. A laboratory processes blank samples alongside real ones through every step, includes a mock community of known composition, and treats any organism that also appears in the blanks with suspicion.

Metagenomics in the clinic

Unbiased sequencing is valuable where the cause of an infection is unknown, because it does not require a hypothesis about which pathogen to test for. In 2024 Benoit and colleagues at the University of California San Francisco reported seven years of results from their laboratory's clinical metagenomic test of cerebrospinal fluid, 4,828 samples tested from June 2016 to April 2023. Five of the paper's 20 authors gave the company Delve Bio as an affiliation. A trade-press report of the study states that the senior author and colleagues co-founded Delve Bio in 2023 and that the company licensed the test's technology from the university.

The San Francisco team compared the test's results with the diagnoses that reviewers reached from the medical charts behind 1,164 of those samples, from 1,053 patients, and in that chart-reviewed subset the test detected 63 % of infections. That figure is the test's sensitivity: the share of true infections that a test detects. Metagenomic sequencing alone identified about one in five of the 220 infections diagnosed in the subset, 48 in all, and every other test run had missed them.

Nanopore sequencing has been tested for speed in research settings: in 2019 Charalampous and colleagues at the University of East Anglia described a nanopore method for bacterial lower respiratory infection that went from sample to result in about six hours. Oxford Nanopore, the sequencer's maker, part-funded the work, and one of the authors was a full-time employee of the company. The team tested the optimized version of the method on 41 respiratory samples and compared its results with culture. Specificity is the companion measure to sensitivity: the share of samples without the target that the test correctly reports as negative. Against culture, the method's sensitivity was 96.6 % and its specificity 41.7 %, and both rose to 100 % once confirmatory tests were applied.

Another test reads microbial DNA from blood rather than from cerebrospinal fluid or respiratory samples: it sequences the cell-free DNA that microbes shed into the bloodstream, and its maker states that the test can identify more than a thousand pathogens. In May 2024 the maker stated that it offered the test as a laboratory-developed test, one that a single laboratory designs and runs itself (Chapter 25). In the same announcement the maker reported that the US Food and Drug Administration, FDA, had granted the test Breakthrough Device designation for suspected lung infections in immunocompromised patients. The designation covers that one intended use, a purpose that the maker states for the test. Breakthrough Device designation is a status the FDA gives some devices before the agency decides on them, and it is neither a clearance nor an approval, two of the decisions that let a test be sold as a device in the United States (Chapter 25).

+ The questionWhat can the vesicles and DNA fragments that cells shed into the blood tell us about tissue we cannot biopsy?

A tube of blood as a sample of the whole body

A tube of blood drawn from the arm holds DNA from the bone marrow, the liver and the walls of blood vessels, and in pregnancy from the placenta. When a cell dies, enzymes cut its DNA into fragments, and some of those fragments reach the bloodstream. Living cells also release membrane-bound packages called extracellular vesicles, or EVs, which carry proteins, RNA and lipids from the cell that made them. Both kinds of material come from almost every tissue, a growing tumor included. A blood draw is therefore a sample, however diluted, of tissues that could otherwise be reached only by a tissue biopsy, and that is the idea behind liquid biopsy.

The same fact sets the difficulty. Material from the tissue of interest is mixed with far more material from everywhere else, and much of it sits at sizes and concentrations near the limits of what instruments can detect.

27 — What a tube of blood holds, by size

DNAProteinMetaboliteReadoutFocal detail

Cell-free DNA: fragments with a fingerprint

Cell-free DNA in plasma is mostly short fragments of about 166 base pairs, and the length has a physical cause. In the nucleus, DNA is wrapped around spools called nucleosomes, with a short stretch of linker DNA between one spool and the next. When a cell dies, enzymes cut the exposed linkers first, so the pieces released are about one spool long. The body clears these fragments from the blood quickly: their half-life there ranges from 15 minutes to several hours, half-life being the measure that Chapter 1 defines. What sits in a tube therefore reflects recent cell death.

Each fragment carries three kinds of information. Its sequence carries any mutation of the cell that released it. Its methylation, the tissue-specific pattern of Chapter 6, matches that of the tissue that released it, so the methylation of many fragments together shows which tissues they came from. In a 2018 study, Moss and colleagues compared the methylation of cell-free DNA pooled from 105 healthy donors with a reference atlas of cell types, and found most of it came from white blood cells and the precursors of red cells, with about 1 % from liver cells. The third kind of information is the fragment's length and end positions, which differ by origin: fetal fragments are shorter, peaking around 143 base pairs, and fragments from a tumor are shorter than healthy ones too.

The first large success was prenatal. Placental DNA enters the mother's blood early in pregnancy, and between 10 and 14 weeks of pregnancy it typically makes up around a tenth of the cell-free DNA, enough to test reliably. A laboratory sequences millions of fragments and counts how many map to each chromosome, and an extra copy of chromosome 21 shows as a small but measurable rise above the expected count. This non-invasive prenatal screening reached the clinic in October 2011, when Sequenom announced that it had launched a test in its own US laboratory for pregnancies at high risk of Down syndrome.

In a 2020 practice bulletin, the American College of Obstetricians and Gynecologists recommended that prenatal screening be offered to every pregnant patient. In 2025 the Society for Maternal-Fetal Medicine recommended that cell-free DNA screening for an extra copy of chromosome 21, 18 or 13 be made routinely available to all pregnant patients, in guidance that the college endorsed. Both documents name cell-free DNA as the most sensitive and specific screen for the common abnormalities of chromosome number (Chapter 21).

Tumor DNA: a detection problem set by the tube

Circulating tumor DNA, the part of the cell-free DNA that comes from a tumor, is the basis of blood tests that choose cancer treatments, monitor response and, increasingly, screen for cancer. In advanced disease it can be a large share of the cell-free DNA. In early disease it is tiny, often below 1 % of fragments.

A 2014 study by Bettegowda and colleagues at Johns Hopkins showed how far so small a share limits detection. The team first found mutations in each patient's tumor and then looked for the same mutations in plasma, with methods based on digital PCR that can detect a single mutant molecule in the DNA from up to 5 mL of plasma. The study's 640 patients formed three groups, and the largest held 410 patients with 18 kinds of tumor. In that group, circulating tumor DNA was detectable in 82 % of the patients whose cancer was at stage IV and in only 47 % of those at stage I, the earliest stage. The 47 % is the figure that the text of the authors' manuscript gives, and a chart in the same manuscript draws it at about 44 %.

The five senior authors owned stock in Personal Genome Diagnostics, a company two of them had co-founded, and the university held patents related to the work. Other authors gave addresses at Life Technologies, Amgen and other companies. The authors note that the shares they found were the most that the amount of plasma collected could give. The limit is set by how few tumor fragments a tube of blood holds, as the example below shows. It is not set by the instrument's analytical sensitivity, its ability to detect a weak signal. Analytical sensitivity is a different quantity from the diagnostic sensitivity of Chapter 13, the share of true cases that a test detects.

How few tumor fragments a tube holds depends on how much cell-free DNA the plasma carries and on the tumor's share of it. In a 2024 study of 862 healthy participants in a cancer screening program in Hong Kong, Malki and colleagues at the Chinese University of Hong Kong measured a median of 7.39 ng of cell-free DNA per mL of plasma. Individual values ranged from 1.61 to 41.01 ng per mL. The paper declared that three of its authors held equity in companies, among them DRA, Take2, Insighta and Illumina, and that several authors were directors of such companies. Two of the three also received patent income from companies that included GRAIL and Illumina, and a patent application had been filed on the study's data.

The methods that work in early disease therefore pool evidence across many positions instead of relying on one. They track many mutations at once, read methylation patterns across thousands of regions, and profile fragment lengths across the genome. Error correction with molecular identifiers, the UMIs of Chapter 12, separates true rare variants from copying errors. In July 2024 the FDA approved a blood test for colorectal cancer screening that combined mutations, methylation and fragment patterns, and in July 2026 it approved a change that makes the test read methylation alone (Chapter 25). Tests that aim to detect many cancers from one draw are further behind, and Chapter 22 examines why the evidence bar for them is so high.

Extracellular vesicles: packages with a return address

EVs are particles enclosed by a lipid bilayer, released by essentially every cell type, and they range from about 30 nm to a micrometer or more. They carry surface proteins from the membrane of the cell that made them and cargo from its interior, including proteins, RNA fragments and metabolites. In principle each vesicle is a small, protected sample of one cell, and that is why EVs attract so much interest, both as biomarkers and as vehicles for delivering drugs.

The International Society for Extracellular Vesicles asks for careful language in its 2024 guideline, MISEV2023, because the biology is easy to overclaim. It treats "small EVs" as an operational term, often used for particles under 200 nm as measured by a given method. It discourages names that imply where inside the cell a vesicle formed unless that origin has been shown.

MISEV2023 also asks a study to report the purity of its EV preparation, because EVs share their size range with other particles in blood. The study should test for at least one protein that marks material other than vesicles. Materials isolated along with the vesicles are called co-isolates, and for blood the guideline asks how far EVs were enriched over the co-isolates that matter to the study, such as platelets, lipoproteins or soluble proteins. Lipoprotein particles, which carry fats and cholesterol, are among the main co-isolates. MISEV2023 notes that all kinds except high-density lipoprotein overlap EVs in size, and that some kinds outnumber EVs in blood by orders of magnitude. Aggregates of protein, and of protein with RNA, also end up in EV preparations.

28 — An EV, and what comes with it

RNAProteinMetaboliteFocal detail

Every isolation method trades purity against yield. Ultracentrifugation pellets particles by density and size, and size-exclusion chromatography separates them by size in a column. Precipitation reagents recover a great deal of material and a great deal of contamination with it. Capture with antibodies against surface markers gives high purity for one subpopulation and excludes the rest. The method chosen therefore decides which EVs a study is actually about.

Measuring vesicles: the detection limit decides the answer

A typical EV is smaller than the wavelength of light, scatters very little of it and carries only a limited number of copies of any one surface protein, so measuring EVs is first a problem of detection limits. Many conventional flow cytometers cannot tell a small EV's scatter from background. In 2018 van der Pol and colleagues reported a multicenter study run through a working group of the International Society on Thrombosis and Haemostasis. In that study, 33 laboratories measured the same samples on 46 flow cytometers, and fewer than half of the instruments detected vesicles in a gate of 600 to 1,200 nm in diameter. The study estimated those diameters from bead calibration and a light-scattering model. Two of its authors declared shares in the company that supplied the study's calibration system.

Other methods reach smaller sizes by different routes. Nanoparticle tracking follows the scattered light of each particle as it drifts and sizes it from how fast it moves, down to about 50 nm, and its fluorescence modes count labeled subsets. High-sensitivity flow instruments and single-molecule fluorescence cytometers count particles by their fluorescent labels instead of their scatter, so detection depends on labeling more than on size. Surface-capture chips and super-resolution microscopes image individual captured vesicles. The consequence is that a measured concentration of EVs depends on the sample, the instrument's detection limit and the analysis settings together.

29 — One sample, three instruments

ReadoutFocal detail

A 2024 bench study shows the point plainly. Kim and colleagues measured portions of one preparation of vesicles from cultured colorectal cancer cells on three instruments: a single-molecule flow cytometer that their University of Washington laboratory had built, and two commercial high-sensitivity flow cytometers, the CellStream from Cytek Biosciences and the CytoFLEX from Beckman Coulter. The authors included researchers at the US National Cancer Institute. Four of the authors declared a financial interest in Pangnostics, which had licensed the single-molecule instrument's technology from the university, and the corresponding author, Daniel Chiu, was one of them.

For vesicles labeled with antibodies to three tetraspanins, surface proteins common on EVs, the single-molecule instrument reported 5.5 × 10⁸ vesicles per mL, the CellStream 2.6 × 10⁷ and the CytoFLEX 7.5 × 10⁶, a spread of about seventy-fold. The vesicles that each counted had median diameters of 57, 83 and 105 nm. Each instrument counted what it could see. From bead calibration, the authors put the smallest vesicle that the CellStream could detect at about 67 nm across and the CytoFLEX's at about 80 nm. They put the single-molecule instrument's limit at 35 nm or less, the size of the smallest vesicle they saw by super-resolution microscopy. They also judged that the beads had set the two commercial limits lower than they really were. When the authors restricted all three data sets to a common range of size and fluorescence, the instruments agreed significantly better, and better still with stricter cutoffs.

Where EV diagnostics stand

EVs carry protected RNA and tissue-specific proteins, which makes them promising candidates as biomarkers. Turning such candidates into validated tests has been slow, largely because of the measurement problems above. One EV-based test for prostate cancer reads RNA from vesicles in a urine sample. In 2019 the company that then made it announced that the FDA had granted the test Breakthrough Device designation (Chapter 25). The company also stated that the test had been added to a clinical guideline. A search of published sources for this guide in October 2026 found no EV-based diagnostic with FDA clearance or approval. Most EV measurement remains a research activity, and the work that will move it forward is largely metrological: calibrated instruments, stated detection limits, reference materials and agreed controls.

So far, the tools for showing that an EV method measures what it claims are mostly reporting rules. MISEV2023 lists the minimal information that a study of EVs should report. In 2020 a working group of vesicle, cytometry and thrombosis researchers published MIFlowCyt-EV, which sets out what an EV flow-cytometry study should report but does not prescribe specific protocols. In 2017 Van Deun and colleagues described EV-TRACK, a knowledgebase that gathers published EV experiments and scores how transparently each one was reported.

As of October 2026 the field still lacks a certified reference material for vesicles, a material whose properties have known values stated on a certificate, so that a method's results can be checked against those values. Chapter 17 places reference materials in the chain of a test and sets out what has been proposed for vesicles, and Chapter 19 describes how a method is shown to work.

+ Part V · The case

Reading the twins.

The combined paper of April 2019 reports that, of the genes whose expression changed during flight in three fractions of the flight twin's blood cells, 91.3 % had returned to normal ranges within six months of landing, and the rest had not. In 2018, NASA's account of the preliminary results had put the two shares at 93 % and 7 %, the figures of the opening story. This part reads the NASA Twins Study with every layer of the guide in hand: first its design and the limits of a study of one pair of people, then layer by layer, then across layers. It answers in full the first clause of the central question: what each omics layer records that the genome cannot.

+ The questionWith the genome held constant, what did combining the layers reveal that no single layer could?

The design, and why it suits the question

On 12 April 2019 the journal Science published the combined results of the NASA Twins Study, a paper by Garrett-Bakelman of Weill Cornell Medicine and 81 co-authors. The study was a case study of a single pair, followed over time: one of a pair of identical twins spent 340 days on the International Space Station, while his brother stayed on Earth. Over 25 months the investigators collected 317 samples from the two brothers before, during and after the flight, among them blood, urine and stool, and they measured the layers of this guide alongside physiology and cognition. NASA's Human Research Program ran the study and led its funding, with further support from US and German research agencies and private foundations. Five of the authors declared roles in companies, as founders, board members, advisers or consultants.

Identical twins start from one fertilized egg, so the ground twin is as close to a genetic control as biology allows. With the inherited genome held constant, every difference the study found had to lie outside the inherited sequence. The investigators also took a step that the chapters on cells anticipated. They separated blood with magnetic beads into four fractions of immune cells, measured RNA in each sorted fraction, and measured methylation in the two fractions that held T cells, one family of white blood cells. Without that step a change in the mixture of cells would have looked like a change inside cells, the averaging problem of Chapter 11.

The design also has obvious limits, and reading it well means keeping them in view. The study followed one pair of people and tested thousands of molecular features at once. Among them were the expression of genes across the genome, 62 cytokines and 4,162 metabolic features in plasma. Cytokines are signaling proteins of the immune system, and the metabolic features are the detected signals of Chapter 10. The authors corrected every result for multiple testing, holding the false discovery rate at 5 % or 1 %. That rate is the expected share of false findings among those reported, so some differences will be chance even after the correction, and Chapter 18 explains how the correction works.

Spaceflight also bundles many exposures together, including microgravity, radiation, confinement, diet and disrupted sleep, and a single case cannot say which one caused a given change. The authors themselves write that, with one person in flight, a change cannot be attributed to spaceflight rather than to coincidence. They call the study hypothesis-generating and write that studies of additional astronauts must complement it.

Layer by layer

The inherited sequence did not change, as expected, but one finding qualifies that statement. The flight twin's blood cells showed more chromosomal rearrangements, such as inversions, during flight, and some were still present after return. Those rearrangements arose during life, in some cells only, and they left the inherited genome as it was. Changes of this kind show that "the genome is constant" is true of the germline, the cells that pass the genome to children, and only approximately true of every cell.

30 — The same genome, measured for 25 months

DNARNAProteinMetaboliteModificationFocal detail

DNA methylation shifted in flight in the two kinds of T cell measured, including at genes involved in immune function and oxidative stress, and largely returned after landing. The paper reports that these genome-wide changes stayed within the range of variation seen in the ground twin, whose methylation changed more over the study and did not fully return to baseline. What set the flight twin apart were changes at the promoters of particular genes, the regions where transcription starts, in groups of genes that did not stand out among the ground twin's changes.

Expression changed for many genes during flight, and most returned to preflight levels within six months while a minority remained changed. Of the genes whose expression changed in three of the sorted fractions, the paper puts the share that returned to normal ranges within six months at 91.3 %.

The telomeres, the protective ends of chromosomes, lengthened in blood cells during flight, by 14.5 % on average, then shortened within about 48 hours of return, leaving more short telomeres than before the mission. In 2019 the authors described the mechanism and the potential consequences of this temporary lengthening as unknown. Protein and metabolite patterns in blood and urine changed in flight, in directions consistent with fluid shifts and physiological stress, and recovered substantially after landing. The composition of the gut community shifted in flight and returned close to its preflight state within six months. The flight twin's cognitive efficiency, measured against the ground twin's, was similar in flight to its preflight level, then declined after return and had not fully recovered at six months. Cognition is not a molecular layer, but it is the observable outcome, or phenotype, that the molecular layers are ultimately trying to explain.

What the combination revealed

Two things were visible only because many layers were measured together. The first is that agreement across layers is evidence, because a shift in immune and stress-response genes seen in methylation, in expression and in proteins is far harder to dismiss as chance than the same shift seen in one data set of thousands of comparisons. A finding in a single layer is therefore a hypothesis, while the same finding in independent layers is an argument.

The second is that the layers moved on different clocks, and not simply the clocks Chapter 1 would suggest. Metabolites, gut microbes and most transcripts responded and recovered quickly, but so did most of the methylation changes. What persisted was a subset of the changes in gene expression, the distribution of telomere lengths, rearrangements carried by long-lived blood cells, and cognition. Timing across layers is information that no single layer holds.

Where the paper sets protein beside RNA, it reports consistency rather than divergence. Collagen proteins rose in the flight twin's urine during flight, and the authors describe that rise as consistent with the raised expression of collagen-related genes in his blood cells. The protein came from urine and the RNA from blood cells, so the comparison shows two layers moving in one direction, and it cannot show how closely one gene's protein follows its own RNA. In general the two need not move together, because of how cells work, and the next section sets out why.

Why RNA does not predict protein

Across genes, the amount of messenger RNA predicts the amount of protein only partly. In an early genome-wide study, published in 2011, Schwanhäusser and colleagues in Berlin measured RNA and protein for more than 5,000 genes in mouse fibroblasts, and found that RNA levels explained about 40 % of the variation in protein levels between genes. They reached that figure on the assumption that their measurements carried no noise. Chapter 1 uses the same study for the half-lives of RNA and protein. In a 2014 reanalysis from the University of California, Berkeley and Lawrence Berkeley National Laboratory, Li, Bickel and Biggin corrected the same data for measurement noise, and for 4,212 genes they put the share considerably higher, at 56 to 84 %.

31 — RNA is not protein

ReadoutFocal detail

Changes in RNA over time within one gene predict changes in its protein much less well, and changes over time are what the Twins Study measured. The reasons are the control points of Chapter 1. Genes differ in how efficiently their RNA is translated, and proteins differ in how long they last. In the mouse cells of the 2011 study, the protein half-lives that could be measured ran from under half an hour to more than 200 hours, so a long-lived protein can stay abundant long after its RNA has fallen. Secreted proteins leave the cell that made them, and a protein's activity can be switched by modification with no change in amount at all. RNA therefore shows what a cell is preparing to do, protein shows what it has, and activity and metabolites show what it is doing, so none of these measurements substitutes for the others.

How layers are combined

Integration methods, which combine the layers in one analysis, fall into three families. Early integration joins all measurements into one table and analyzes it together, which is simple but lets the largest data set dominate. Late integration analyzes each layer separately and combines the conclusions, which respects each layer's statistics but misses relationships between them. Intermediate integration looks for a small number of hidden factors that explain variation across all layers at once, and it is widely used for studies with many samples.

Genetic variation adds a fixed reference point, because inherited variants do not change with the state of the cell. When a genetic variant is associated with the level of an RNA, a protein and a metabolite in the same direction, and with a disease, the chain from code to phenotype can be traced step by step across a population. The large population studies of Chapters 9 and 21 measure several layers in the same people for that reason.

What each layer records that the genome cannot

The central question asks first what each omics layer records that the genome cannot. The twins supply the answer in miniature, and the chapters before this one supply the mechanism. The table's last column gives the typical time over which each layer changes; single molecules can be faster or slower, as the range of protein half-lives above shows.

Each layer is a different record of state laid over the same code. The layers differ in cost and difficulty for a physical reason, whether their molecules can be copied and read by pairing, and that answers the central question's second clause. They differ in meaning because each sits at a different control point between instruction and function. Parts VI to IX take up the third clause: what it takes to turn one of these measurements into a result someone can act on.

+ Part VI · The test as a system

How a measurement becomes a result.

On 2 February 2026 an amended version of the FDA's quality rule for device makers took effect under a new name, the Quality Management System Regulation, which incorporates the international standard ISO 13485:2016. The rule governs how device makers design, make and check what they sell, and it applies to the makers of diagnostic test kits. Part VI looks at an omics test in the same way, as a designed system. Chapter 16 takes the instrument and its subsystems, Chapter 17 the chain from the blood tube to the signed report and the people who own each step, Chapter 18 the reading of a result, and Chapter 19 the evidence that a test measures what it claims before anyone relies on it. Chapter 20 takes what keeps that evidence true once the test is in use: control of suppliers and of changes, complaints and corrective action, reports of wrong results, field actions and post-market follow-up.

+ The questionWhat happens inside the instrument between loading a sample and writing a data file?

Three instruments, one architecture

A short-read sequencer is a cabinet of pumps, a heated glass flow cell, lasers, a camera and a computer. A mass spectrometer and a flow cytometer look nothing like it from the outside, yet seen as systems the three share one architecture. Each has a way of handling the sample and delivering it to the measurement, a physical conversion that turns molecules into something detectable, a sensor, detection electronics, a controller that keeps every step in time, and software that turns detector counts into an answer. Instrument engineers will recognize the signal chain behind every measuring instrument, sense → condition → digitize → compute → connect → decide, and an omics instrument is that chain applied to a biological signal.

32 — Two instruments, one architecture

ReadoutFocal detail

The subsystems of a short-read sequencer show the architecture at work. Fluidics pump reagents across the flow cell in a fixed sequence, while the flow cell holds the chemistry at a controlled temperature. In each cycle, lasers excite the dyes incorporated in that cycle and a camera images the surface, so that image processing can locate every cluster and measure its intensity in each color, and base calling then converts those intensities into letters, each with a quality score (Chapter 4).

A mass spectrometer has the same shape: pumps and a chromatography column deliver peptides over time, electrospray converts them into ions, ion optics and mass analyzers sort the ions, and a detector counts them, after which acquisition software writes spectra and search software turns them into identities. In both instruments no single component explains the result. The coordination of the components does.

Timing limits read length and quantification

In a sequencer, every cycle must add one base to every strand, image the whole surface and remove the blocking group before the next cycle begins. If a small fraction of the strands in a cluster miss an addition, they fall behind the rest, a problem known in sequencing chemistry as phasing, a different sense of the word from the phasing of variants in Chapter 5. Strands that add two bases in one cycle run ahead instead, which is called prephasing. Either way the cluster's signal becomes a blend of the current position and its neighbors, and because phasing and prephasing accumulate cycle by cycle, base quality declines along a read and read length is limited. The controller's job is therefore to hold flow, temperature and timing tightly enough that each cycle is as close to identical as the chemistry allows.

In a mass spectrometer, timing decides whether a peptide can be quantified at all. Each peptide emerges from the chromatography column over a few seconds, and to measure its amount the instrument must sample it several times across that peak. Every other ion the instrument is also trying to measure competes for the same seconds, so each measurement added to the instrument's cycle lengthens the cycle and leaves fewer samples of each peak.

The data-conversion chain

The most useful way to look at any omics instrument is to follow one fact as each module re-represents it. In sequencing, a DNA fragment becomes a cluster of copies, the cluster becomes a flash of light in each cycle, the flash becomes pixel intensities, and the intensities become a base call with a quality score. The calls then become a read, the read an alignment, and alignments a count or a variant call. In proteomics, a protein becomes peptides, the peptides become ions, the ions become a spectrum, and the spectrum becomes a peptide match with a score. Matches then become an inferred protein, and signal becomes a quantity.

33 — The data-conversion chain

DNAProteinReadoutFocal detail

At no point does the instrument measure the quantity of interest directly. It measures a proxy and converts it six or seven times over, which means that the result is only as trustworthy as the weakest conversion. Each conversion has a characteristic failure mode, and the quality metrics that accompany an omics result exist to watch those conversions one by one.

Controls travel the whole chain

A metric describes one conversion, while a control tests all of them at once. A control is a known quantity passed through every step alongside the samples, so that anything lost or distorted along the way shows up as a difference between what went in and what came out. Each kind of instrument has its own kind of control: sequencing runs carry a control genome whose sequence is known, RNA-seq experiments can include synthetic RNAs at known concentrations spanning the expected range, targeted mass spectrometry adds isotope-labeled copies of each measured peptide, and cytometers run beads of certified brightness.

The logic is the same in every case: the only way to know how well a chain of conversions worked is to push something known through it. The instrument, though, is only one link in a longer chain, which runs from the blood draw to the signed report. Chapter 17 follows it, with the people who own each step and the faults beyond the instrument that controls must also catch.

+ The questionWho owns each step between drawing the blood and signing the report, and where can one fault turn into a wrong answer?

Seven links between a tube and a report

The moment a tube of blood fills for a cell-free DNA test, the sample inside it begins to change. Until the tube is centrifuged, the white cells in it release their own long genomic DNA into the plasma, where it mixes with the short fragments the test is meant to read (Chapter 14). The US National Cancer Institute, NCI, therefore names centrifugation within 2 hours of the draw as optimal in its evidence-based practice for cell-free DNA. That limit is for a standard tube that holds EDTA, a chemical that keeps the blood from clotting. NCI accepts up to 4 hours at room temperature or 24 hours at 4 °C, and up to 3 days at room temperature for the one brand of stabilizing tube it names. None of those hours passes inside an instrument. They pass in a clinic, in a courier's box and in a laboratory's receiving area, handled by people who never see the sequencer.

Chapter 16 followed a measurement through the instrument. A test is longer: it is the whole chain from the patient's arm to a signed report, and the instrument is one link in it. The specimen is collected in a tube of a stated type, and pre-analytical handling moves, spins and stores it. Library or assay preparation turns the extracted molecules into something the instrument can read (Chapter 4). The instrument and its primary analysis turn the signal into reads or spectra (Chapter 16). The pipeline, the software that aligns, counts and calls, turns reads into results. Interpretation decides what a result means for one patient, and the report carries that decision to the clinician.

Each of these seven links hands its output to the next across an interface, and a fault introduced at any link reaches the report unless a later link is built to catch it. The NASA Twins Study of Chapter 15 drew on several investigator teams, each measuring its own layers, so its combined result rested on several such chains at once. A clinical test runs one chain, but that chain still crosses several organizations, each answering for different links.

Five kinds of party own the links. Reagent makers supply the tubes and the preparation kits, an instrument maker supplies the instrument and its primary software, and software makers, commercial or open source, supply the pipeline's tools (Chapter 23 names the makers of instruments and kits as of October 2026). The laboratory runs all of them and joins them into one test, and the clinician orders the test and acts on its report. None of the documents this chapter cites assigns all of these owners. The division follows from what each party makes and from what US law asks of the laboratory.

The law that governs US clinical laboratories is CLIA, the Clinical Laboratory Improvement Amendments, which Congress passed in 1988. Under it, a laboratory that runs an FDA-cleared or approved test without change must show that it obtains accuracy, precision and a reportable range comparable to the maker's before it reports patient results. A laboratory that builds its own test, or modifies a cleared one, must establish the test's performance itself (Chapter 19). The maker of a cleared kit therefore answers for the performance it established, and the laboratory answers for reproducing that performance, or for establishing it anew when the laboratory has built or changed the test.

34 — The test as a chain, with its owners and formats

ReadoutFocal detail

Two documents written for test developers treat the test as a chain in the same way. One is FDA's final guidance of April 2018 on sequencing tests for inherited disease, still current as of October 2026. It asks a developer to specify and document every test element, from instruments and software to consumables and reagents, together with the specimen types, the regions read, and the controls and reference materials. ISO is drafting a standard for human DNA sequencing workflows, ISO/DIS 25379-1, which runs from pre-examination through library preparation, sequencing and analysis to reporting, and which applies to diagnostic examinations including laboratory-developed tests. Its ballot as a draft international standard opened on 16 September 2026, so as of October 2026 it is a draft and not yet a published standard.

The specimen changes before the laboratory receives it

Everything that happens to a specimen before the analytical test makes up its pre-analytical phase, which ISO's standards call pre-examination. It is the link the laboratory controls least, because much of it happens before the specimen arrives. Its faults also change the molecules themselves, which the instrument then measures faithfully. Genomic DNA released into a waiting tube adds fragments that carry none of a tumor's mutations, so the tumor's share of the cell-free DNA falls and a low-level variant becomes harder to find.

NCI's document records that the yield of genomic DNA rose after delays of 4, 6, 24 and 48 hours in different studies. It also names two checks for that contamination. One compares how well a long and a short stretch of DNA amplify, since cell-free fragments are too short to hold the long one, and the other reads the fragment sizes on an electropherogram.

The tube is the first specification, because tube types hold different anticoagulants, chemicals like EDTA that keep blood from clotting. For cell-free DNA, NCI names EDTA tubes as optimal and acid citrate dextrose as acceptable, advises against heparin and citrate, and recommends a stabilizing tube when a delay of more than 2 hours is expected. Processing follows in two spins. The first runs at 800 to 1,600 g, multiples of the force of gravity, for 20 minutes. The second runs at 14,000 to 16,000 g for 10 to 20 minutes and removes the remaining cells and debris that would otherwise release genomic DNA.

For the separated plasma, NCI recommends 3 hours or less at 4 °C as interim storage, then up to 3 months at −20 °C or up to 9 months at −80 °C. It accepts −80 °C for long-term storage. Three cycles of freezing and thawing degraded cell-free DNA in the studies it reviewed. NCI's advice on tubes is written for cell-free DNA, and a reference material made for another analyte, the substance a test measures, can use another tube. The US National Institute of Standards and Technology, NIST, issues a reference plasma for metabolites, described at the end of this chapter, and that plasma was collected in lithium heparin, one of the anticoagulants NCI advises against for cell-free DNA.

35 — The pre-analytical clock for cell-free DNA

ReadoutFocal detail

Hemolysis, the bursting of red cells in the tube, is a further fault, and its timing depends on the tube. NCI's document reports hemolysis in one brand of stabilizing tube after as little as 24 hours, and in another brand after 14 days at room temperature. A stability limit therefore holds for the brand that was tested rather than for every tube of its type.

ISO's standards set requirements and recommendations for the same phase. ISO 20186, published in three parts in 2019, covers venous whole blood for cellular RNA, genomic DNA and cell-free DNA from plasma. ISO 20184 covers frozen tissue, and ISO 20166 tissue fixed in formalin and embedded in paraffin, each with parts for RNA, proteins and DNA. The users that ISO names for the cell-free DNA part are laboratory customers, test developers and manufacturers, biobanks, research organizations and regulators, which covers nearly every party in the chain. As of October 2026 ISO lists that part as current, confirmed in December 2024.

Each interface has a format and an owner

Library preparation, described in Chapter 4, comes next: a kit from a reagent maker, run by the laboratory's staff to the kit's protocol or to the laboratory's own modification of it. The instrument and its primary analysis follow, as Chapter 16 described, and at the boundary between the instrument and the pipeline the result becomes a file. Illumina's guide to its BCL Convert software describes one such step on the company's own instruments. The software converts the binary base-call files, which hold the data cycle by cycle, into FASTQ files, which hold them read by read, and it sorts the reads by sample.

Each FASTQ record holds a read's sequence and a quality score for every base. The format had no formal definition until Cock and colleagues published one in December 2009, and they documented at least three incompatible variants in use. A program therefore had to know which variant it was reading to report the quality scores correctly, and the same risk arises wherever one party writes a file and another reads it.

Aligned reads are written as SAM, a tab-delimited text format, or as BAM, the same data compressed. Each alignment line holds 11 mandatory fields, among them the read's position on the reference, its mapping quality and its base qualities, which SAM encodes as the Sanger variant of FASTQ does. Variant calls are written as VCF, the variant call format. Each data line of a VCF file describes one position in the genome in 8 fixed columns, among them the reference and alternative bases, a quality and a filter field that reads PASS when the position passed every filter, followed, when the file holds genotypes, by a format column and one column for each sample. A work stream of the Global Alliance for Genomics and Health maintains these specifications, and as of October 2026 the current VCF version, 4.5, is dated 25 February 2026.

Mass spectrometry has its own interface file. mzML is the format of the Human Proteome Organization's Proteomics Standards Initiative for raw spectrometer output, developed by researchers, instrument vendors and software developers together. It replaced the earlier mzData and mzXML formats, now deprecated, and as of October 2026 its version 1.1.0, released on 1 June 2009, is still the current one. Groups outside any one company maintain SAM, BAM, VCF and mzML, and no standards body maintains FASTQ, whose incompatible variants included the ones that Solexa and Illumina software wrote. What the laboratory owns is the choice, tool by tool, of which format version each program writes and reads, because its pipeline joins tools from several makers.

Interpretation and the report are the last links, and the laboratory owns both. Interpretation turns a call into a statement about a patient, such as whether a variant explains the condition the clinician asked about. The report carries that statement to the clinician, and FDA's 2018 guidance asks that it also list the test's limitations, including any regions that failed sequencing. What happens after the report belongs to the clinician.

Single faults and the controls against them

The first two faults in this section are single: each has one cause, one effect and a control designed against it, and the first of them has been measured in detail. Laboratories pool many libraries on one flow cell to share its capacity, so each library carries a sample index, a short known sequence in its adapters, and after the run the reads are sorted back by sample, a step called demultiplexing. The sample index does for whole samples what the cell barcode of Chapter 12 does for single cells. Index hopping is the appearance of a read carrying the index of a library it did not come from, so that demultiplexing assigns the read to the wrong sample.

In 2018 Costello and colleagues at the Broad Institute's sequencing service measured how often index hopping happens. They analyzed two years of their own production data, more than 150,000 exomes and more than 57,000 genomes sequenced with two indexes per library, and ran a mixture experiment with human and bacterial libraries. On Illumina's HiSeq X, HiSeq 4000 and NovaSeq instruments, which use patterned flow cells, the swap rate ranged from 0.2 to 6 % of reads and averaged about 1 %. Illumina describes the surface of such a flow cell as nanowells at fixed, evenly spaced locations, in which clusters grow by a method it calls exclusion amplification.

The rate depended on the library as well as the instrument. PCR-free genome libraries, prepared without amplification, typically swapped about 3 % of reads, and libraries made with amplification about 0.25 %. In means over eight pools and flow cells, the PCR-free genome libraries swapped 3.01 % of reads on HiSeq X and 4000 instruments and 4.85 % on a NovaSeq, against 0.13 % on a MiSeq, an instrument outside the patterned group. The authors suggest that PCR-free libraries, whose yield is low, carry more free adapter for each fragment, that is, more adapter molecules left unattached after preparation. The instruments they measured date from 2014 to 2018, and the paper reports no rates for later models.

Illumina, the instruments' maker, states in a 2018 white paper that index hopping on patterned flow cells is 2 % or less, and 1 % or less on non-patterned ones. The Broad's measured rates for PCR-free genome libraries on HiSeq X, 4000 and NovaSeq instruments were above the first of those figures. Illumina names free adapters as a cause of higher rates and recommends removing them, storing libraries at −20 °C and sequencing pools promptly.

The control the Broad measured checks both indexes of every read. With unique dual indexes, each library carries an index at each end and no two libraries in the pool share either index, so a read whose two indexes belong to different samples is set aside as undetermined. On the same HiSeq X data, sorting by both indexes lowered the mean contamination of the PCR-free genomes from 0.89 % to 0.13 % of reads. A read whose two indexes both swap into another valid pair still passes, and the authors estimate such double swaps at about 0.01 %.

36 — Index hopping and unique dual indexes

ReadoutFocal detail

A sample swap moves an entire result to the wrong patient, and every run metric stays normal, because each swapped sample was measured correctly. The control is an identity check against a second, independent measurement of the same person. In 2013 Pengelly and colleagues at the University of Southampton and the National Genetics Reference Laboratory in Wessex described a panel of 24 single-nucleotide polymorphisms, or SNPs, positions at which people commonly carry one of two different letters, and every commonly used exome kit captures all 24. The laboratory genotypes the same 24 positions with a separate assay and compares the two sets of genotypes, and a mismatch exposes a swap. The authors tested the panel's power on 3,780 genomes from the 1000 Genomes and UK10K projects and in 20,000 simulated replicates. Applied to 48 exomes, it found one reciprocal switch, between two samples in adjacent wells.

Lots and batches, sets of reagents made together or samples processed together, cause faults of another kind, because they shift many samples at once. Costello's group saw the swap rate differ between flow cells for the same pool of libraries, with the setup of the amplification, the reagents and the flow-cell lots. The group also found cross-contamination between neighboring wells in one vendor's index plates, a fault that arrived with a reagent. Chapter 18 takes up these batch effects, the shifts that laboratory conditions, reagent lots and personnel leave across many samples, and shows how they are found in the data and why a balanced design is their control.

Reference materials test the chain from where they enter

Each control above watches one fault. A reference material tests many links at once: a material whose answer is known in advance passes through every link from the point where it enters the chain, and the difference between the known answer and the report measures the error of those links. Chapter 16 introduced the control as a known quantity passed through every step. NIST issues such materials for three of this guide's layers, and its documents for them state what each value can support. NIST's material for genomes is purified DNA and its material for metabolites is plasma already separated from the blood cells, so neither passes through the tube or the spins, and the pre-analytical limits of this chapter need checks of their own.

For genome sequencing, NIST's Genome in a Bottle work produced RM 8398, DNA from one large growth of the cell line GM12878, which came from a woman of Utah and European ancestry. NIST describes it as intended for validation, optimization and process evaluation, and for assessing variant calling from genome sequencing. Its values for single-nucleotide variants, small insertions and deletions and positions that match the reference are information values, characterized but not certified. NIST states that they "cannot be used to establish metrological traceability", the documented link from a result to a recognized reference. Its benchmark regions cover about 91 % of GRCh37 and about 85 % of GRCh38, two versions of the human reference genome, and NIST lists sister materials for a family trio of Ashkenazi ancestry, two parents and their child, and for a person of Chinese ancestry.

For RNA the known material is added to each sample instead of being run beside it. NIST's SRM 2374 holds 96 DNA templates, chosen from a set of 176 and developed with the External RNA Controls Consortium, from which a laboratory transcribes spike-ins, synthetic RNAs added in known amounts to every sample. In a 2011 study, Jiang and colleagues at the National Institutes of Health, NIST and Cold Spring Harbor Laboratory added a pool of 96 synthetic RNAs to RNA-seq experiments. The pool spanned about 20 doublings in concentration, a millionfold. Reads rose in proportion to input across the range the method detected. The scatter between replicates was larger than random sampling alone predicts, however. Each RNA's count was also biased by its length, by its GC content, the share of its letters that are G or C, and by the priming that starts its copying. Distortions of this kind are what a spike-in exists to expose.

For metabolites, NIST's SRM 1950 is frozen plasma pooled from 100 fasting adults, equal numbers of men and women aged 40 to 50. NIST intends it for validating methods that measure metabolites such as fatty acids, electrolytes, vitamins and hormones. About 34 of its analytes carry certified values and about 55 more carry values without certification. The material therefore checks a targeted measurement of those analytes, and it offers no certified profile for the many features an untargeted run reports (Chapter 10).

For extracellular vesicles, NIST's list holds no reference material as of October 2026. A Ghent University group wrote in 2019 that no universal reference material existed for separating, calibrating or normalizing vesicle measurements, and proposed recombinant fluorescent vesicles as a trackable biological reference. Participants in two workshops of the International Society for Extracellular Vesicles stated in a position paper of 2020 that reference materials for vesicle size, concentration, refractive index and the abundance of surface proteins are required. Chapter 14 showed what the gap costs: without calibration and stated detection limits, vesicle counts from different instruments cannot be compared.

+ The questionHanded an omics report, how do you tell a real finding from noise?

What a differential table reports

A differential result is a table that compares two groups of samples feature by feature, and most omics studies end in one. Here a feature is anything the study measured, whether a gene, a protein or a metabolite, in a wider sense than the signals of an LC-MS run that Chapter 10 calls features. For each feature the table reports four things. It gives how large the difference between the groups is, as a fold change, and how surprising that difference would be if nothing were really different, as a p-value. It adjusts that p-value for the number of features tested. And it gives how abundant the feature was, because features present in small amounts are measured with more noise.

The 7 % of the opening story was a share of such a list: of the changes in gene expression seen during spaceflight, the share that had not returned to preflight levels six months after landing. Reading any such report means reading all four quantities together and knowing what each can and cannot say. The first, the fold change, is a ratio. If a gene averages 200 normalized counts in treated samples and 50 in controls, the fold change is 200 divided by 50, or 4, and the reverse comparison, 50 against 200, gives 0.25 (Chapter 7 explains normalized counts).

On a ratio scale those two look unequal, although they describe the same size of change in opposite directions. Counting the change in doublings makes them symmetric, which is why reports give the log₂ fold change: a fourfold rise is two doublings, a log₂ fold change of 2, and a fall to a quarter is two halvings, or −2. The rule of thumb practitioners carry follows from that: a log₂ fold change of 1 is a doubling, −1 is a halving, and every further unit is another factor of two. A value of 3 is therefore an eightfold increase, while a value of 0.2 is a 15 % increase, the kind of change that can be statistically significant and biologically unimportant.

Significance, and the price of testing everything

A p-value answers a narrow question: if there were truly no difference, how often would chance alone produce a difference at least this large? At the conventional threshold of 0.05, about one test in twenty will come out significant when nothing is going on. For one planned test that is an acceptable risk, but across 20,000 genes the same threshold fails badly, because 1,000 genes would be expected to pass by chance alone even if the treatment did nothing. Omics analysis therefore controls the false discovery rate, or FDR, which Chapter 8 introduced for the matching of spectra to peptides: the expected share of false findings among everything declared significant. The standard procedure comes from a 1995 statistical methods paper by Benjamini and Hochberg: it ranks the p-values and compares each with a threshold that rises with its rank.

The volcano plot

The standard picture of a differential result is the volcano plot, which places every feature by its log₂ fold change across the page and by its significance up the page. Significance is plotted in tenfold steps of the p-value, so that 0.001 sits one step above 0.01 and stronger evidence sits higher, and the resulting shape gives the plot its name. Features in the upper corners changed a lot and reliably, while the crowd at the bottom did not change detectably.

37 — Anatomy of a volcano plot

RNAFocal detail

Two regions deserve suspicion rather than excitement. High up near the middle sit features with tiny changes and very small p-values, because with enough samples even trivial differences become highly significant. Far out at the sides but low down sit features with large changes and weak evidence, usually because they had few counts and therefore noisy estimates. The findings worth pursuing combine a change large enough to matter with evidence strong enough to trust.

Quality metrics decide whether any of it holds

The biology in a report can be interpreted only after the measurement itself has been shown to work, so a careful reader checks the quality metrics first. Each layer has its own quality metrics, the ones Chapter 16 tied to specific conversions, and the table lists them with what a problem looks like. Two rows of the table rely on the coefficient of variation, or CV, the spread of repeated measurements as a percentage of their mean, measured on a pooled quality-control sample that mixes a little of every sample and is run at intervals through the batch. For untargeted metabolomics, a 2018 review of quality-control practice by Broadhurst and colleagues gives a CV of 20 % or, for some sample types, 30 % as the usual limit for keeping a feature.

When the batch is the biggest signal

Samples are rarely processed in one go. They are extracted on different days, run on different plates and sequenced on different flow cells. Each set of samples processed together is a batch, and each batch leaves a small systematic shift on every measurement. With thousands of features, those small shifts add up to a signal that can be larger than the biology. A 2010 review by Leek and colleagues described how such batch effects lead to wrong conclusions when they line up with the comparison a study is making.

Principal component analysis shows the problem. It finds the directions along which the samples differ most and plots the samples along them. In a well-behaved experiment the leading components do not line up with processing batches, and they may reflect the groups being compared or other biology such as sex or donor. When the leading components separate the processing days instead, the batch dominates. If every treated sample was processed on one day and every control on another, the two are perfectly confounded, and no analysis can separate them afterwards.

38 — When the batch is the biggest signal

ReadoutFocal detail

The remedy therefore lies in the design of the experiment more than in any correction afterwards. In a sound design each group is spread evenly across batches, the run order is randomized, and pooled quality-control samples are included at intervals through every batch, because statistical batch correction works only when the design gives it something to work with. A batch effect is one fault of a longer chain that runs from the blood tube to the report, and Chapter 17 describes the others, such as swapped samples and contamination, together with the controls that catch them.

Reading a report row by row

The table below is an extract in the form of a typical differential expression report, with illustrative values and a plain-language reading of each row. Two of its rows carry the traps of the volcano plot. Gene 2 rose by only 13 %, yet its abundance and its adjusted p-value make the change reliable, so it is real and probably unimportant. Gene 3 changed about tenfold, but the estimate rests on a mean count of about nine reads per sample, and its p-value of 0.004 becomes 0.21 once it is adjusted for the number of genes tested. The last row is as important as any gene. A spike-in is one of the controls described in Chapters 16 and 17: a synthetic molecule added at the same amount to every sample. It should show no change, and if it did, every other row would be in doubt.

+ The questionBefore the first result reaches a patient, what must the team behind a test be able to show, and how does it judge whether the risk that remains is acceptable?

A test is proven against a claim stated exactly

In August 2014 a consortium coordinated by the FDA reported one kind of measurement that the RNA sequencing it had tested did not deliver: an accurate absolute amount of each transcript. The Sequencing Quality Control project, known as SEQC or MAQC-III, measured reference samples at several laboratory sites on three platforms, Illumina's HiSeq, Life Technologies' SOLiD and Roche's 454, and produced more than 100 billion reads. Its authors included staff of Illumina, Life Technologies and other companies that make gene-expression products. Neither RNA sequencing nor the microarrays it was compared with gave accurate absolute measurements of expression. Relative expression, the ratio between two samples, was accurate and reproducible across sites and platforms "if specific filters are used", and the consortium wrote that "Measurement performance depends on the platform and data analysis pipeline".

The finding divides one claim into two. A claim about how much a gene's expression changed between two samples passed the consortium's comparison, on stated platforms, with a stated pipeline and stated filters. A claim about how many copies of each transcript a sample holds did not. Proving a test therefore starts before any experiment, with a claim stated exactly enough that an experiment could fail it: the specimen, the analyte or substance measured, the kind of result, the platform and the pipeline. Everything the team later shows is shown against that claim, and against nothing wider.

For a device maker in the United States, the route from such a claim to a finished design is set by the Quality Management System Regulation, or QMSR, the current form of 21 CFR Part 820. FDA issued the final rule on 31 January 2024, the Federal Register published it on 2 February 2024, and it took effect on 2 February 2026 in place of the Quality System Regulation. The QMSR incorporates by reference the requirements of ISO 13485:2016, the international standard that sets quality-system requirements for device makers, which ISO reviewed and confirmed in 2025. FDA states that a certificate of conformance to ISO 13485 "will not exempt a manufacturer from an FDA inspection".

Clause 7.3 of ISO 13485, on design and development, sets out what US practice calls design controls, the name of the corresponding section of FDA's earlier Quality System Regulation. Under the QMSR it applies to devices in class II and class III, the two higher of FDA's three risk classes (Chapter 25), and to the class I devices that are automated with computer software or named in a list in the rule. These requirements bind the maker of a device, such as a kit sold to laboratories. A US laboratory that builds its own test answers for it under CLIA, whose legal frame Chapter 25 sets out together with what became of FDA's 2024 rule on such tests. A laboratory can apply the same discipline of written requirements and kept evidence to a test it builds.

Every requirement points to its evidence

FDA's training material on the QMSR sets out what clause 7.3 asks, step by step. A plan documents the stages of the work, its reviews, its responsibilities, its resources and its traceability. Design inputs, the requirements, must be complete, unambiguous, able to be verified or validated and free of conflict, and they must "Account for all pertinent risk management outputs". Design outputs must meet the inputs and contain or reference their acceptance criteria. For an omics test the outputs include its protocols, its pipeline, its specifications and its labeling, the documents and software that embody the design. Reviews check the work at planned points.

Verification shows that each output meets its input, and validation shows that the finished design meets its users' needs and its intended use. Validation runs on representative product to a documented plan and is completed before release. Transfer then moves the design into manufacturing, and every later change passes through change control before it is implemented: the change is reviewed for its effects, and the studies it touches are run again as the review requires (Chapter 20). All of it is recorded in the design and development file, kept for every device type or family, which must "include or reference records that demonstrate conformity". Traceability makes the file usable: each requirement points to the outputs that meet it and to the records that prove it, so a reader can start from any claim and reach its evidence.

Chapter 17 introduced FDA's final guidance of April 2018 on sequencing tests for inherited disease, with the test elements it asks a developer to specify. The same guidance shows what design inputs look like for an omics test, and it covers germline tests only, not tests of tumors or of cell-free DNA. Beside the elements Chapter 17 listed, the developer specifies the methods for DNA quantity and quality and for library preparation and enrichment. The developer then sets thresholds for accuracy, precision and the limit of detection, the smallest amount of a variant the test reliably finds, and run-quality thresholds for average and minimum depth, uniformity of coverage and base quality. The guidance gives no number for any of those run-quality thresholds, neither a depth such as 30× nor a share of bases at Q30, so the developer sets each one and justifies it.

39 — From claim to evidence, one trace

Focal detail

The specimen requirement belongs among the inputs as well. Chapter 17 showed that the tube, the time to centrifugation and the storage conditions change what the instrument receives, so a claim made for blood drawn into one tube and processed within one window holds only for specimens handled that way. The input then names those limits, the output is the specimen instruction, and verification has to show that the result holds at the edges of the window the instruction allows.

A wrong result becomes dangerous when someone acts on it

Some design inputs come from the claim, and others from an analysis of what could go wrong. The standard for that analysis is ISO 14971:2019, which ISO confirmed in a review of 7 March 2025 and lists as current as of October 2026. Its catalogue entry describes a process for the risk management of medical devices across their life cycle, with in vitro diagnostic devices and software included. The QMSR does not incorporate ISO 14971 by reference, and FDA's training on design and development does not name the standard. The training states instead that design inputs must account for all pertinent outputs of risk management, the process that ISO 14971 sets out.

ISO's guidance on applying the standard, ISO/TR 24971:2020, carries an annex for in vitro diagnostic devices, and ISO registered a project on 6 July 2026 to replace that guidance with a technical specification, ISO/TS 24971-1, which as of October 2026 is at an early stage. ISO's page for the project notes that ISO 14971 does not require its process to sit inside a quality management system, and that applying it can meet some of the risk requirements of ISO 13485.

ISO 14971's terms, as Health Canada and FDA reproduce them, separate the source of harm from the harm itself. A hazard is a potential source of harm, and a hazardous situation is a circumstance in which people, property or the environment are exposed to one or more hazards. Harm is injury or damage to the health of people, or damage to property or the environment. Risk combines the probability of occurrence of harm with the severity of that harm. FDA's introductory training on risk reproduces the chain the standard draws between these terms: a hazard, a sequence of events, a hazardous situation and harm. The probability of occurrence of harm is then the product of two probabilities, that the hazardous situation occurs and that it leads to harm.

For an omics test, which works on a specimen already taken, the path to harm runs through the result that someone acts on. This guide sorts wrong results into three kinds: a false negative, which misses a condition that is present; a false positive, which reports one that is absent; and a delayed result, correct but too late for the decision it was meant to inform. Regulators' documents use broader words. IMDRF, the international forum of device regulators, speaks in its 2025 document on software risk of "information-related hazards" and of harm that includes "reduction of effectiveness". The EU's guidance on post-market surveillance speaks of "expected erroneous results", a term Chapter 20 takes up. Whatever the words, this guide reads ISO 14971's definitions to mean that a wrong result becomes a hazardous situation only when it reaches a decision about a patient.

Index hopping, measured in Chapter 17, can be traced through every step of this chain. The hazard is the presence of reads from another patient's library in a pooled run. The sequence of events is that some of those reads carry a variant, the pipeline calls it above its filters, and interpretation reports it as relevant. The hazardous situation is a false-positive result in front of a clinician who must decide on treatment, and the harm is a treatment or a procedure the patient did not need. Each step of the sequence is a place where a control can stop it.

40 — A wrong result on the path to harm

Focal detail

Chains such as this one are found by systematic analysis, and two general methods each have an international standard. A failure modes and effects analysis, or FMEA, starts from the parts and steps of a design or a process, asks how each might fail and follows each failure to its effects. IEC's catalogue entry for the 2018 edition of IEC 60812 states that the purpose of the method "is to establish how items or processes might fail", and that "Failure modes can be prioritized to support decisions about treatment." Treatment there means what is done about a risk, not the care of a patient. A fault tree analysis works in the opposite direction to an FMEA, from one undesired event back through the events and failure modes that could combine to produce it, and IEC 61025, of 2006, gives guidance on it. Both standards are in their current editions as of October 2026, and both come from IEC's committee on dependability rather than from medical-device work.

FDA's training on risk management under the QMSR lists a process FMEA, which "Assesses manufacturing process risks", among a maker's risk documents. On the hopping chain, an FMEA of the pooling step would list index hopping as one way that step can fail and follow it to a false call. A fault tree drawn down from a false-positive report would set hopping beside a sample swap, the other single fault of Chapter 17, as events that can each lead to it.

A risk is accepted only against criteria set in advance

Finding and estimating the risks does not yet decide whether a test may be released with them. FDA's training on risk management under the QMSR names the risk documents in which that decision is made. The risk management plan "Outlines risk management scope, responsibilities, and criteria", so the criteria for accepting a risk are fixed before any risk is estimated, and each estimated risk is then judged against them. A risk that fails the criteria needs a control.

ISO 14971 ranks the controls. FDA reissued its guidance on human factors on 3 August 2026, and the guidance lists them "in order of preference and effectiveness": inherent safety by design, then protective measures in the device itself or in the manufacturing process, then information for safety. On the hopping chain, the design measure is unique dual indexes, which remove most hopped reads before any call is made. A protective measure is a check that stops a result, such as a run that must pass its quality thresholds before its calls are released. An identity fingerprint that must match is the same kind of check against the sample swaps of Chapter 17. Information for safety is the report's statement of the test's limitations, which FDA's 2018 guidance asks for. Each control then becomes a requirement in its own right and enters the trace with its own verification.

What a control leaves is the residual risk. FDA's human factors guidance reproduces the definition of ISO 14971: "Risk remaining after risk control measures have been implemented." The guidance states that "some residual risk will remain" in any device, and that "The remaining residual risk is then evaluated for acceptability", which means a second judgment against the plan's criteria. The guidance is written about the risks of using a device, and the same reasoning applies to a wrong result. On the hopping chain, the foreign reads left after dual indexing can still produce a false call, so that residual risk is judged with the run's quality thresholds and the report's limitations in place.

The last judgment covers all the residual risks together. FDA's training on risk management under the QMSR names two documents for it: a benefit-risk analysis, made when residual risks remain after the controls, that "Justifies why benefits outweigh residual risks", and a risk management report that "Justifies acceptability of individual and overall risk". The overall residual risk, all the residual risks taken together, is therefore weighed against the benefit of the test. FDA's human factors guidance asks a submission in the same way to "describe how the overall benefits of using the device outweigh the residual risks."

41 — From an estimated risk to an accepted residual risk

ReadoutFocal detail

Software that measures has a life cycle of its own

In an omics test, software does much of the measuring. Base-calling software turns images or currents into letters (Chapter 16), and the pipeline aligns, counts and calls (Chapter 17). The international standard for the life cycle of device software is IEC 62304, in force as of October 2026 in its edition 1.1, the 2006 text with its 2015 amendment. It sets requirements for developing and maintaining software that is itself a medical device or an integral part of one. A normative reference is another standard whose requirements a standard takes in, and the only one IEC 62304 names is ISO 14971, so the software process of IEC 62304 depends on risk management carried out to ISO 14971.

IEC 62304 sorts software into three safety classes, A, B and C, and it sets requirements in several clauses for software of unknown provenance, or SOUP. The Johner Institute, a company that sells regulatory training and consulting, reproduces the standard's definition: SOUP is software already developed and generally available but not developed for incorporation into the device, or software developed earlier for which adequate records of its development are not available. IEC's committee has drafted a design specification for a second edition. That edition would apply "to the life cycle of health software", not only to device software, and would replace the three classes with two levels of process rigor, so that "Current Class A is equivalent to level I, Classes B and C are intended to be level II". As of October 2026 the second edition is a proposal, and edition 1.1 is the standard in force.

FDA issued final guidance on the content of premarket submissions for device software functions on 14 June 2023, and the guidance is current as of October 2026. It asks for one of two documentation levels. The Enhanced level applies where a failure or flaw of the software "could present a hazardous situation with a probable risk of death or serious injury", and the Basic level applies otherwise. The level decides what a maker submits, not how the software is built, and for its development practices a maker may submit a declaration of conformity to the FDA-recognized version of IEC 62304.

FDA calls a component outside the maker's full control off-the-shelf software, "a generally available software component used by a device manufacturer for which the manufacturer cannot claim complete software life cycle control". On that wording, an open-source aligner or variant caller would fit the definition. For a germline sequencing test, FDA's 2018 guidance treats the pipeline as part of the test. It asks the developer to document "software versions and traceability, reference sequence assembly", to include "the finalized bioinformatics pipeline" in the end-to-end validation of the test, and to document every modification, "including software updates and other modifications to the bioinformatics pipeline".

A pipeline validated this way is locked: every tool, version, parameter and reference is fixed before validation, so the pipeline that runs on patients' samples is the one that was tested. For a minor change, and only for one, the guidance allows "comparing results from the new pipeline to the existing test pipeline".

IMDRF's 2025 document adds a principle for judging the risk of any medical device software. Where possible, it suggests setting "the probability of software failure to 1", so that the analysis assumes the software will fail and judges the consequence. It also suggests categorizing that risk "prior to implementation of risk controls", before any control is credited.

Analytical validation, measure by measure

The word validation also has a laboratory sense, different from the design validation above. Analytical validation shows that the test measures its analyte correctly, and CLIA's regulation lists the measures. Before reporting patient results from a test it built or modified, a US laboratory must establish, as applicable, accuracy, precision, analytical sensitivity, analytical specificity including interfering substances, the reportable range, the reference intervals and any other characteristic the test requires. For an unmodified cleared or approved test, the laboratory verifies the maker's accuracy, precision and reportable range and checks the reference intervals against its own patients. CMS, which runs the CLIA program, reviews that validation at its routine survey every two years, after testing has started, and its 2013 answers on laboratory-developed tests state that the program "does not address the clinical validity of any test".

CLSI, a standards organization for laboratory medicine, publishes a consensus document for each measure in the table below, which lists the documents as CLSI's catalogue describes them in October 2026.

CLSI's pages list EP06, EP07, EP09, EP12 and EP17 as recognized by FDA, and they do not state whether FDA recognizes EP05's new edition or MM09. EP05, EP06 and EP09 are written for quantitative procedures, while EP12 covers a binary call such as a variant present or absent. EP07 was written for clinical chemistry, so applying it to the interferents of an omics test, such as hemoglobin released by hemolysis, is the developer's own extension. MM09 introduces bioinformatics and defers detailed guidance on it to a future CLSI document.

For a germline sequencing test, FDA's 2018 guidance names its own measures. Accuracy is reported as positive percent agreement, the share of known variants the test calls; negative percent agreement, the share of known reference positions the test leaves uncalled; and technical positive predictive value, the share of the test's calls that are true. The developer sets a threshold for each, on both the point estimate, the value the study measured, and the lower bound of its 95 % confidence interval, the lowest value of the range that the data support at that level of confidence. Precision is tested across "multiple samples, runs, reagent lots, days, and operators", and the limit of detection for each variant type is the level at which "at least 95% of positive calls" are obtained. A reference material such as RM 8398 can supply the known variants, within the regions its benchmark covers (Chapter 17).

The three figures answer different questions. Negative percent agreement is counted over every reference position the test reads, and those far outnumber the positions that carry a variant, so a few false calls barely move it. Technical positive predictive value is counted over the calls the test makes, so every false call lowers it, and it shows directly what share of the variants in a report are true.

The measures CLIA lists and FDA's guidance names are all analytical: each asks whether the test measures its analyte correctly. Whether the result tracks a condition in the intended patients, and whether acting on it helps them, are the clinical rungs of the evidence ladder in Chapter 22, and no analytical study answers them.

The laboratory's own quality system surrounds these studies. The College of American Pathologists, CAP, an accreditation body, added 18 requirements for sequencing tests to the 2014 edition of its molecular pathology checklist. Aziz and colleagues, led from CAP, described them in a report published online in 2014 and in print in 2015. The requirements cover the laboratory bench work and the bioinformatics, and the committee behind them weighed documentation, validation, confirmatory testing, the monitoring of upgrades, version traceability, variant interpretation and reporting, and the storage and transfer of data. CAP's current checklist is not public, so the count describes the 2014 edition.

ISO 15189:2022 is the fourth edition of the international standard for quality and competence in medical laboratories, and it applies to point-of-care testing as well. In a 2026 comparative review, Gupta and Agarwal, writing from a Delhi institute, find that the 2022 edition makes risk management a requirement throughout every process, more strongly than the 2012 edition did. In India, NABL, the National Accreditation Board for Testing and Calibration Laboratories, posted its notice of the move to the 2022 edition on 29 December 2022. It required its specific criteria for medical laboratories, NABL 112A, to be in place by 31 December 2025, checked at each laboratory's next assessment. How each jurisdiction uses these systems is the subject of Part IX.

The evidence assembled before release has to stay true afterwards, as reagent lots, instruments, software and the tested population change, and Chapter 20 follows how: through the control of suppliers and of changes, complaints and corrective action, reports of wrong results, field actions and post-market follow-up.

+ The questionWhat has to keep happening for a test that was proven once to stay proven while people rely on its results?

Suppliers, controlled in proportion to risk

On 27 July 2026 the FDA received notice of a change to a blood test for colorectal cancer that it had approved three days earlier. The test's maker reported changes in a supplier's manufacturing processes for critical reagents, using a 30-day notice, the route that Chapter 25 describes for changes to an approved device's manufacturing methods. The FDA recorded the notice as a process change at a supplier, not as a change to the test's design, and accepted it on 26 August 2026, 30 days after receiving it. Yet the studies behind the approval had used reagents from the supplier's earlier process, and kits made after the change could carry reagents from the new one.

Chapter 19 set out how a test is proven against a claim, with traced requirements, a locked pipeline and analytical studies, and the proof describes the test as it stood when the studies ran. After release the test's inputs keep changing. Suppliers change their reagents and software, the maker changes its own design, and laboratories and clinicians report complaints and wrong results that no validation study saw. Since 2 February 2026 the FDA has inspected device makers under a compliance program that names the parts of a quality system that deal with these changes. Among them are outsourcing and purchasing, complaint handling, corrective action, preventive action, and product and process changes. Together they form a loop that runs for as long as the test is in use. These duties fall on the maker of a device, and Chapter 25 shows how a US laboratory's own test reaches patients under CLIA instead.

An omics test is assembled largely from other makers' products. Its tubes and preparation kits, its instrument and primary software, and the tools of its pipeline come from different makers, the owners that Chapter 17 named. A maker that sells the test as a device buys many of these parts, and each company it buys from is a supplier. Work that an outside party does for the maker is what ISO 13485 calls an outsourced process. The standard, which the QMSR incorporates, requires the control of both to be proportionate to risk, and FDA's training on the regulation reproduces its words. The criteria for evaluating and selecting suppliers must be "proportionate to the risk associated with the medical device", and the controls over an outsourced process "proportionate to the risk involved and ability of external party to meet requirements".

The same rule of proportion applies after the purchase. In the standard's words as the FDA reproduces them, the extent of verification activities "shall be based on the supplier evaluation results", and a failure to meet purchasing requirements is "addressed with the supplier proportionate to the risk". The QMSR's own text no longer carries a section on purchasing. FDA's comparison of the old and new rules lists the purchasing controls of the old regulation among the requirements now incorporated by reference. The preamble to the final rule states that the QMSR "requires that manufacturers of finished devices apply an ongoing" assessment of their suppliers, informed by risk.

A supplier's change reaches a proven test through faults of the kind Chapter 17 described. There, the rate of index swapping varied between flow cells with the reagents and the flow-cell lots, and one vendor's index plates arrived with cross-contamination between neighboring wells. A supplier that changes how it makes an enzyme or an index plate can move such a rate while every step in the laboratory stays the same. Incoming checks and the test's own controls are then what detect the shift. The evaluation of each supplier sets how closely its products are checked on arrival. Software written outside the maker enters the test in the same way, and Chapter 19 showed that an update to an open-source tool in the pipeline is a change to the test.

A change is reviewed before it reaches patients

Changes also start with the maker. A new reagent formulation, a new instrument model, a new version of the pipeline or a new specimen tube each alters something that the evidence of Chapter 19 measured. FDA's training on the QMSR sets out that ISO 13485 requires such changes to be controlled before they are implemented, which is the change control of Chapter 19. Within that process the change review judges what a change affects, and the training reproduces what clause 7.3.9 of the standard asks of it: the "review of design and development changes shall include evaluation of the effect of the changes on constituent parts". For an omics test those parts include the reagents, the instrument's settings and the pipeline. A change to a library kit is therefore judged by its effect on the quality of the run, on the pipeline's filters and on the calls that reach the report.

The trace of Chapter 19 sizes the work that follows. Because each design input is linked to the outputs that satisfy it and to the studies that proved it, a change can be followed to the inputs it touches, and the studies behind those inputs are run again. A new lot of an enzyme from a supplier that has passed its evaluation may need only a comparison with the previous lot on a reference material. A change to the variant caller means that accuracy and the limit of detection are shown again for every type of variant it calls. For a minor change to the pipeline of a germline sequencing test, FDA's 2018 guidance allows the new pipeline's results to be compared with those of the existing one (Chapter 19). A change that could alter whether the test still meets its intended use, such as a new specimen type, returns the team to validation.

Some changes must also go to whoever assessed the test before it was sold. In the United States, the notice at the opening of this chapter reached the FDA before the supplier's change was made, and Chapter 25 shows a change to another approved blood test, to its classification model, that needed the FDA's approval before it was made. In the European Union a test's assessor is its notified body, a conformity assessment body designated under the In Vitro Diagnostic Regulation, or IVDR, and Chapter 26 sets out which tests need one. Annex IX of the IVDR obliges a maker to tell its notified body in advance about substantial changes it plans to make to its quality system or its range of devices. In the annex's words the maker "shall inform the notified body", and the body records its approval as a supplement to the certificate. When the notified body has assessed a device's technical documentation, changes need that body's approval "where such changes could affect the safety and performance of the device", and the approval again takes the form of a supplement to the certificate. The same annex requires the notified body to audit each maker without notice at least once every five years.

India's Medical Devices Rules, 2017, put the licensing authority's approval before a major change. Under Rule 26(iii) a license holder "shall obtain prior approval" from the central or state licensing authority, whichever applies, "before any major change as specified in the Sixth Schedule". Chapter 27 sets out which authority licenses which test. The authority is to approve or reject the change within 45 days, and without a reply "such change shall be deemed to have been approved". Minor changes, which the same schedule lists, are reported within 30 days, and an import license carries the same duties with 60 days for the authority's answer. CDSCO, the Central Drugs Standard Control Organisation, publishes a compilation of the Rules with their amendments to October 2022, and the wording above is taken from it. CDSCO's addendum of 13 March 2026 to its answers to frequently asked questions about IVDs states that each change to the version of approved software needs an application for a post-approval change.

A complaint opens an investigation

The users of a test report what goes wrong with it as complaints, and each complaint is a possible sign that the evidence no longer holds. For an omics test a complaint may be a laboratory reporting runs that fail their quality thresholds, or a clinician reporting a result that disagreed with another test, with a repeat of the same test or with the patient's course. The QMSR fixes what the maker's record of a complaint must hold. The record gives the device's name, the date the complaint was received, the unique device identifier or other identifiers, the complainant's contact details, and the nature and details of the complaint. It also holds any corrective action and reply, and the records of review and investigation, with a justification when no investigation was done.

FDA's page on mandatory reporting adds that a maker keeps complaint files and procedures "for receiving, reviewing, and evaluating complaints". Complaints that must be reported to the FDA are kept "in a separate portion of the complaint file or otherwise clearly identified". The page is dated 27 March 2025, before the QMSR took effect. Once a complaint is on file, the investigation looks for its cause. The preamble to the QMSR's final rule names clauses 8.2.2, 8.5.2 and 8.3.1 of ISO 13485 as the ones that address investigations of complaints. Clause 8.5.2, on corrective action, covers evaluating "any need for actions to ensure that nonconformities do not recur", where a nonconformity is a failure to meet a requirement, such as a result outside the test's specification.

Clause 8.5.3 adds preventive action, taken on a potential problem before it occurs, such as a weakness found in one lot that may also lie in lots not yet used. FDA's training reproduces how the standard scales both: "Corrective actions shall be proportionate to the effects of the nonconformities encountered", and "Preventive actions shall be proportionate to the effects of the potential problems." The preamble states that the FDA "has incorporated the corrective action and preventive action requirements of ISO 13485 by reference into the QMSR".

The IVDR draws the line between the two kinds of action in another place. It defines corrective action as "action taken to eliminate the cause of a potential or actual non-conformity or other undesirable situation", so its term also covers the potential problems that ISO 13485 places under preventive action. A maker that works under both therefore maps its procedures to both sets of words. The IVDR's terms for action in the field, the subject of this chapter's last section, are built on its definition of corrective action.

For an omics test a corrective action can be a change: an acceptance test for incoming lots, a tighter threshold for run quality, a revised filter in the pipeline, or a new line among the report's limitations. Each then passes through change control, with the change review of the previous section and the re-verification that the review calls for.

The risk documents of Chapter 19 are updated with what the investigation found, such as a hazard that the risk analysis had not listed or a probability that it had set too low. ISO 14971 places this work in its clause on production and post-production activities. A 2022 summary by the Johner Institute, the regulatory consultancy of Chapter 19, describes that clause as asking the maker to collect information from the installation, use and maintenance of the device, among other sources, and to review it for hazards not previously considered and for risks that were wrongly estimated or are no longer acceptable. The investigation also settles two further questions, whether the event must be reported to a regulator and whether tests already in use need action, and the next two sections take them in turn. The changed test then returns to use.

42 — The loop that keeps a proven test proven

ReadoutFocal detail

Harm through a decision, and the clocks for reporting it

A diagnostic test harms in a way that an implant or a pump does not. It acts only on a specimen already taken, so a wrong result injures no one directly, and the harm comes through what a clinician or a patient decides on the result (Chapter 19). FDA's guidance on reporting and the IVDR's definitions both name that path, and each of the three jurisdictions sets its own clocks. Each clock for reporting an event starts on a day set by what the maker has learned, whether or not the investigation has yet found a cause.

In the United States the Medical Device Reporting regulation, 21 CFR Part 803, sets out who reports what to whom. A maker reports deaths, serious injuries and malfunctions to the FDA within 30 calendar days of becoming aware of an event. The limit is 5 work days when the FDA designates the event, or when the event "requires remedial action to prevent an unreasonable risk of substantial harm to the public health". An importer reports deaths and serious injuries to the FDA and to the maker, and malfunctions to the maker only, within 30 calendar days. FDA's page lists user facilities as hospitals, ambulatory surgical facilities, nursing homes and outpatient diagnostic or treatment facilities. A user facility reports a death to the FDA and the maker, and a serious injury to the maker, within 10 work days.

The FDA explains the words that start these clocks in its final guidance on reporting by manufacturers, issued on 8 November 2016 and not withdrawn as of October 2026. For a 30-day report, and for a 5-day report that the FDA has requested, a maker becomes aware when any of its employees becomes aware of information that "reasonably suggests that an event is required to be reported", so that clock starts on the day the first employee learns of such information. A 5-day report on an event that needs remedial action has its own starting day. For that report the maker becomes aware when an employee in a particular role learns, from any information including a trend analysis, that the action is needed. That day can come later than the day the first employee heard of the event.

The guidance describes that role in a single clause. It names an employee "with management or supervisory responsibilities over persons with regulatory, scientific or technical responsibilities", and the clause goes on "or whose duties relate to the collection and reporting of adverse events". The wording does not settle whether those duties belong to the employee or to the persons supervised, and the regulation's own definition could not be read for this guide. On the first reading, the clock starts when either a supervisor or an employee who handles adverse events learns that the action is needed. On the second, it starts only when a supervisor does.

A malfunction is "the failure of a device to meet its performance specifications or otherwise perform as intended", and FDA's guidance adds that "Performance specifications include all claims made in the labeling for the device." For a test the guidance is direct: "we consider the failure of a diagnostic device to be reportable when it indirectly causes or contributes to, or is a factor in, a death or serious injury, or would be likely to, if the failure were to recur."

The IVDR names harm through a decision in its definitions. An incident is "any malfunction or deterioration in the characteristics or performance of a device made available on the market". The definition includes use error due to ergonomic features and any inadequacy in the information the maker supplies, and it also includes "any harm as a consequence of a medical decision, action taken or not taken on the basis of information or result(s) provided by the device". A serious incident is one that "directly or indirectly led, might have led or might lead to" a death, a temporary or permanent serious deterioration in a person's state of health, or a serious public health threat. Harm that follows when a clinician withholds treatment on a wrong result is therefore an incident in the regulation's own words, and a serious incident when that decision led, might have led or might lead to a death or a serious deterioration in health.

Article 82 sets the clocks for serious incidents. Makers report a serious incident "immediately after they have established" a causal relationship with the device, or that such a relationship is reasonably possible, and "not later than 15 days after they become aware of the incident". The limit is 10 days for a death or an unanticipated serious deterioration in a person's state of health, and 2 days for a serious public health threat. One exception is particular to tests: "expected erroneous results which are clearly documented and quantified in the product information" are left out of these reports. On that wording, the performance figures in a test's instructions also separate the wrong results that the maker reports one by one from those it counts in trends.

Trend reporting, under Article 83, covers non-serious incidents and expected erroneous results together. The Medical Device Coordination Group, whose guidance documents interpret the regulation, describes the duty in guidance of December 2025, which is not legally binding. The guidance asks the maker to watch these events for a "statistically significant increase in the frequency or severity" that could significantly affect the balance of benefit and risk, and to define the methods, thresholds and observation period in its surveillance plan. After a serious incident, Article 84 requires the maker to "perform the necessary investigations" without delay, including a risk assessment of the incident and of any corrective action taken in the field, which the last section of this chapter describes. The maker cooperates with the competent authorities, the member states' regulators, which evaluate the risks themselves.

India's Rules use one phrase for every kind of device. Under Rule 26(ii) the license holder informs the state or central licensing authority, whichever applies, "of the occurrence of any suspected unexpected serious adverse event and action taken thereon including any recall". It must do so "within fifteen days of such event coming to the notice of licence holder". Unlike the IVDR's definition of an incident, the words name no path to harm through a result. A newspaper, a law firm and a compliance-software firm reported in May and June 2024 that a CDSCO circular of 15 May 2024 addressed license holders on reporting adverse events from their devices through the Materiovigilance Programme of India, MvPI, the national program that collects reports of adverse events from medical devices. The three summaries differ on whether the circular requires such reports or urges them, and on whether it covers all adverse events or only suspected unexpected serious ones. None gives a time limit or a form or says whether the circular names IVDs, and the circular itself could not be read for this guide.

43 — Reporting deadlines after a maker learns of a death: the US, the EU and India

ReadoutFocal detail

Field actions, and follow-up for the test's whole life

When an investigation shows that tests already in use put patients at risk, the maker acts on them where they are. In the United States, 21 CFR Part 806 governs two kinds of action. A correction is the repair, modification, adjustment, relabeling, destruction or inspection of a device, including patient monitoring, without moving it, and a removal moves the device from its point of use. A maker or importer reports a correction or removal to the FDA when it is made to reduce a risk to health posed by the device, or to remedy a violation of the law that may present a risk to health. The report is due "within 10-working days of initiating such correction or removal". Improvements that reduce no risk to health, market withdrawals, routine servicing and stock recoveries are exempt from the report.

The maker keeps records of all corrections and removals for 2 years beyond the device's expected life. On these definitions, an update to a cleared test's pipeline made to reduce a risk to health would be a correction. FDA's compliance program of February 2026 ties ISO 13485 to the two US rules. A maker meets clause 8.2.3 of the standard, on reporting to regulators, by notifying the FDA of complaints that meet the criteria of Part 803, and it handles its advisory notices to users "in accordance with the requirements of 21 CFR 806".

On 20 March 2026 Foundation Medicine began to recall 163 units, in three lots, of a sequencing reagent for its FoundationOne CDx test, the profiling test of 324 genes that Chapter 21 describes. FDA's record of the recall gives the reason: the reagent might increase phasing, in the sense of Chapter 16, and so produce false-positive calls of small insertions or deletions at splice sites, the edges of exons, in the BARD1 gene. The company asked the sites that had received the lots to quarantine the units and return them. On 22 April 2026 the FDA classified the action as a Class II recall, a class that belongs to the recall and not to the test, which is a device of the FDA's class III (Chapter 25). As of October 2026 the record gives the cause as under investigation by the firm, and it does not say how BARD1 calls already reported were handled.

The IVDR calls a field action a field safety corrective action, a corrective action that a maker takes "for technical or medical reasons to prevent or reduce the risk" of a serious incident. The message that announces one to users is a field safety notice. A US correction is reported after it starts, while a field safety corrective action is reported "in advance of the field safety corrective action being undertaken", except in urgent cases. Article 84 requires the notice to "explain, in a clear manner, without understating the level of risk", to "clearly indicate all the actions to be taken by users", and to be entered in the regulation's electronic system, "through which that notice shall be accessible to the public." Four modules of EUDAMED, the EU's database for devices, became mandatory on 28 May 2026, and the module for vigilance, the reporting of incidents and field actions, is not among them (Chapter 26).

India's Rules define a recall as "any action taken by its manufacturer or authorised agent or supplier to remove the medical device from the market or to retrieve the medical device from any person to whom it has been supplied". They give three grounds for one: the device is hazardous to health, it fails to conform to a claim its maker made about its quality, safety or efficacy, or it does not meet the requirements of the Act and the Rules. For a test, the second ground ties a recall to the performance its maker claimed. Under Rule 26(vi), when the licensing authority directs it, the license holder withdraws the rest of a lot from sale and recalls the units already issued "so far as practicable".

Field actions deal with one problem at a time, and post-market surveillance watches the test as a whole. India's Rules define post-marketing surveillance as a systematic process for collecting and analyzing information from devices placed on the market. The IVDR builds a system around the same idea in Articles 78 to 81, which the Medical Device Coordination Group's guidance of December 2025 sets out. A maker plans, runs and updates a surveillance system whose plan is part of the test's technical documentation. It prepares a surveillance report for a test in class A or B and a periodic safety update report for one in class C or D, the two higher of the IVDR's four classes of risk (Chapter 26).

Post-market performance follow-up is required under the IVDR to confirm, over the device's expected lifetime, that its safety, performance and clinical evidence remain valid and its balance of benefit and risk acceptable. The coordination group's guidance calls it "a continuous process that updates the performance evaluation", the evaluation of Chapter 26 with its three elements. Data from tests in use therefore return to the evidence that was assessed before sale, and the next change review starts from that updated evidence rather than from the studies of the first release.

+ Part VII · From signature to decision

From discovery to routine use.

In 2010 a set of cancer trials at Duke University was halted. The trials had used gene-expression signatures to assign patients to chemotherapy, and independent statisticians could not reproduce the results on which the signatures rested. Part VII asks what separates a signature like those from the omics tests that already inform decisions. Chapter 21 maps where omics decides today, from prenatal screening and cancer treatment to rare disease, prescribing and manufacturing, and Chapter 22 sets out the evidence a signature needs, from locked validation to proof that acting on it helps patients, before it becomes a test someone can act on.

+ The questionWhere is omics already changing decisions, not just papers?

A working definition of "in use"

In September 2026 investigators reported that their randomized trial in England of a blood test for many cancers had missed its primary endpoint (Chapter 22). As of October 2026 the test is still awaiting a decision by the FDA, and it remains a candidate for use.

A technology is in use when a validated test built on it informs a decision routinely, is paid for, and appears in clinical guidelines or manufacturing specifications. By that definition, omics is further along than its reputation suggests in a few areas and further behind in most. The routine uses are also narrower than the word implies: none of them reports on a whole layer. Each reports a validated subset of one, chosen because it answers a specific question, even where a whole genome is sequenced, as in the rare-disease diagnosis this chapter describes.

44 — Where omics already decides

DNARNAProteinMetaboliteModificationFocal detail

Screening from a blood sample

One of the most widespread clinical uses of genome sequencing is also one of the oldest: counting cell-free DNA fragments in maternal blood to screen for fetal chromosome abnormalities, a kind of test first launched in 2011. It rests on the counting property of short-read sequencing described in Chapter 5 and on the fragment biology of Chapter 14. In a practice bulletin of 2020, the American College of Obstetricians and Gynecologists recommended that prenatal screening be offered to all pregnant patients regardless of age or risk, and it named cell-free DNA as the most sensitive and specific screen for the common abnormalities of chromosome number. The Society for Maternal-Fetal Medicine wrote that bulletin with the college, and in 2025 it issued guidance of its own, which the college endorses as of October 2026. That guidance recommends that cell-free DNA screening for an extra copy of chromosome 21, 18 or 13 be made routinely available to all pregnant patients, and it again names cell-free DNA as the most sensitive and specific screen for those common abnormalities.

Screening for cancer from a blood sample is newer. As of October 2026 the FDA has approved three blood tests for colorectal cancer screening. The first, approved on 12 April 2016, detects methylated DNA of the Septin 9 gene in plasma, and it is approved for adults aged 50 and older at average risk who have been offered the recommended screening tests and have declined them. The FDA approved the second in July 2024, and on 24 July 2026 it approved the third, which looks for molecular signals of colorectal cancer in cell-free DNA from plasma. Both recent tests are approved for adults aged 45 and older at average risk, and the FDA's record for the 2026 test states that a positive result should be followed by colonoscopy. Chapter 22 gives the performance figures of the 2024 test and explains why screening healthy people raises the bar for evidence, and Chapter 25 describes how such tests are approved.

Choosing and following cancer treatment

Cancer treatment is where omics has become most embedded, in three forms. Tumor genome profiling reads tens to hundreds of genes in a biopsy to find mutations that match a targeted drug. In 2017 the FDA approved a broad profiling test covering 324 genes as a companion diagnostic, a test essential for the safe and effective use of a particular drug. Two blood-based profiling tests followed in 2020, and in 2024 the FDA approved a distributable kit covering 517 genes, with claims across tumor types.

Expression signatures measure a panel, a fixed set of genes, in a tumor to estimate the risk that the cancer will return. In 2007 the FDA granted a De Novo request for a 70-gene signature, the first multigene test of its kind. The FDA uses a different verb for each of its main routes: it approves a test of the highest risk class, grants a De Novo request for a novel test of low or moderate risk, and clears a test shown to be equivalent to one already on the market (Chapter 25).

Two large trials then changed practice. The first, TAILORx, was a randomized trial run by the ECOG-ACRIN cancer research group with funding from the US National Cancer Institute. It enrolled 10,273 women with early breast cancer that was hormone-receptor-positive and HER2-negative and had not spread to the lymph nodes. Of the women enrolled, 6,711 with intermediate scores on a 21-gene assay were assigned at random to endocrine therapy alone or with chemotherapy. In 2018 Sparano and colleagues reported that endocrine therapy alone was not inferior to adding chemotherapy, although women aged 50 or younger with scores of 16 to 25 saw some benefit from chemotherapy.

The second, MINDACT, was a randomized trial that enrolled 6,693 women with early breast cancer and used the 70-gene signature to assess genomic risk. In 2016 Cardoso and colleagues reported that many women at high clinical risk but low genomic risk could forgo chemotherapy, judged by survival without distant metastasis at five years. Two of the authors were then at Agendia, the company that makes the 70-gene signature.

The third form is methylation. In brain tumors the methylation of a single gene, MGMT, is routinely tested because it predicts response to a standard chemotherapy (Chapter 6).

In November 2025 the FDA proposed moving nucleic-acid-based companion diagnostics for cancer drugs from its highest risk class to a lower one, governed by special controls, which are requirements written for that kind of test. This guide reads the proposal as a sign that the category has matured, and as of October 2026 it is still a proposal. A parallel order did become final: it covers companion diagnostics based on in situ hybridization, which marks a gene in a tissue section with a probe that pairs to it, and the order took effect in September 2026 (Chapter 25).

On 15 May 2026 the FDA approved a companion diagnostic that tests for molecular residual disease, the traces of cancer that remain after treatment and can be detected only by molecular tests. The test is designed from each patient's own tumor and detects circulating tumor DNA in plasma (Chapter 14). It is intended to identify patients with muscle-invasive bladder cancer who may benefit from treatment with the drug atezolizumab. Four days later the FDA approved a blood-based panel with a much larger footprint than the 2020 tests, covering small variants in 741 genes.

Payment, the second condition of the definition at the start of this chapter, is settled differently in each jurisdiction. As of October 2026, Medicare covers sequencing-based tests for patients whose cancer is recurrent, relapsed, refractory, metastatic or advanced at stage III or IV, under a national coverage determination first effective in March 2018. Among other conditions of that national coverage, the test must be an FDA-approved or cleared companion diagnostic for the patient's cancer and must be run in a CLIA-certified laboratory (Chapter 25). The same determination allows Medicare's regional contractors to cover a sequencing test for such a patient when the FDA has not approved or cleared it. The Medicare contractor Palmetto GBA designed and runs MolDX, a program that sets the coverage, coding and pricing of molecular diagnostic tests for several of Medicare's regional contractors. Palmetto GBA's 2019 manual for the program states that CMS, the federal agency that runs Medicare, has directed MolDX to follow the ACCE criteria that Chapter 22 describes. As of October 2026, Medicare pays for most laboratory tests at the rates of its Clinical Laboratory Fee Schedule.

In the European Union, payment for a test can depend on a health technology assessment, an evaluation of a medicine, device or test that informs a health system's decisions on paying for it. A market-access consultancy states that companion diagnostics are often the subject of such assessments because of their high cost and novelty. Regulation (EU) 2021/2282 on health technology assessment has applied since January 2025. It provides for joint clinical assessments, allows tests of the highest risk class to be selected for them and leaves pricing and reimbursement to each member state. In India, a member of Parliament stated that the public insurance schemes, Ayushman Bharat among them, cover only the inpatient costs of cancer treatment and leave diagnostic tests to be paid for out of pocket.

Rare disease, from diagnosis to newborn screening

For children with suspected genetic disease, sequencing has become the first test rather than the last, and some of the evidence for that comes from studies that also included adults. Meta-analysis, the pooling of results across studies, puts the diagnostic yield of genome sequencing at 34 % to 41 %, compared with about 10 % for the chromosomal microarrays it replaces. In 2018 Clark and colleagues pooled 37 studies of 20,068 children in one such meta-analysis and found a yield of 41 % for genome sequencing. Their estimate rests on 374 of those children, in seven studies, and the paper's senior author was then at an institute that offers rapid genome sequencing. In 2023 Chung and colleagues pooled 161 studies of 50,417 children and adults with rare disease and found a yield of 34 % for genome sequencing, with a 95 % confidence interval of 30 % to 38 %.

45 — What the first genetic test finds

ReadoutFocal detail

In a pilot published in 2021, the investigators of England's 100,000 Genomes Project sequenced the genomes of 4,660 children and adults from 2,183 families with undiagnosed rare disease. They reported a genetic diagnosis for 25 % of the probands, the affected family members through whom the families entered the study, and about a quarter of the diagnoses had immediate consequences for care. In intensive care, rapid genome or exome sequencing typically returns results within a few days, and in record cases within hours. In a 2024 review, Kingsmore and colleagues pooled 44 studies of critically ill children with diseases of unknown cause, and they reported a genetic diagnosis in about 37 % of the children and a change in management for about a quarter. Kingsmore was also the senior author of the 2018 meta-analysis, and when the review was published he and his co-authors were at the same institute, which offers rapid sequencing.

Newborn screening by tandem mass spectrometry, a targeted metabolomics test on a dried blood spot, has been routine since the late 1990s and screens every baby in many countries for dozens of inherited metabolic conditions. Genome sequencing at birth is now being tested at scale. Genomics England runs the Generation Study with the NHS and aims to screen the genomes of up to 100,000 newborns for over 200 rare genetic conditions that can be treated in early childhood. It reported in October 2025 that 25,000 babies had enrolled. In 2024 the GUARDIAN study in New York reported results from its first 4,000 newborns. Columbia University Irving Medical Center and NewYork-Presbyterian ran the study with funding from the drug maker Sanofi, the sequencer maker Illumina, the genetic testing company GeneDx and the Columbia Precision Medicine Initiative.

Infection, drugs and manufacturing

Unbiased metagenomic sequencing is used for hard-to-diagnose infections, as Chapter 13 described, and pathogen genome sequencing has become a routine tool for tracking outbreaks and antimicrobial resistance. Pharmacogenomics uses a person's genotype to choose a drug or its dose. The Clinical Pharmacogenetics Implementation Consortium maintains guidelines for gene–drug pairs. The FDA also keeps a table of pharmacogenomic biomarkers in drug labeling, the official prescribing information that comes with each drug, and the table runs to several hundred entries.

In biopharmaceutical manufacturing, omics methods serve as quality-control instruments. Mass spectrometry characterizes therapeutic proteins and identifies residual host-cell proteins individually. It also monitors glycosylation, the sugar chains attached to a protein, as a critical quality attribute, which is a property that should stay within set limits to ensure the drug's quality (Chapter 10). Sequencing confirms the identity and stability of the cell lines that make the drug.

Population cohorts: the research base for the next decade

Large cohorts that pair omics with health records are where most future clinical uses are being discovered. In 2025 the UK Biobank Whole-Genome Sequencing Consortium reported whole-genome sequences for 490,640 of the biobank's participants. The US All of Us Research Program's August 2026 data release includes more than 535,000 short-read genomes and over 14,000 long-read genomes, with proteomics and RNA sequencing on thousands of samples. GenomeIndia's 10,000 genomes give Indian research its own reference (Chapter 4). UK Biobank is also measuring proteins in 600,000 samples, with the full data due in 2027, and the expansion will link genetic risk to circulating proteins at a scale not seen before (Chapter 9).

+ The questionWhy do many omics discoveries never become tests, and what does it take for one that does?

Discovery and validation are different experiments

With 20,000 random features measured in 40 patients, a model can sort the patients perfectly into their two groups and then do no better than chance on new ones. An omics discovery study works in exactly that situation. It measures thousands of features, whether genes, proteins or metabolites, in tens or hundreds of samples, and it looks for a pattern, or signature, that separates two groups. When a study has many more features than samples, some combination of features will separate any two groups perfectly, even when every measurement is random, which means that a model fitted to such data can learn the noise of that particular set of samples as easily as the biology.

46 — How a signature fits noise

ReadoutFocal detail

The only reliable defense is data the signature has never seen. Like a pipeline in Chapter 19, the signature is fixed, or locked, with every step of its processing and its decision threshold, and it is then tested on independent samples. Those samples are ideally collected elsewhere, processed at a different time, and drawn from the population in which the test would be used. Many omics signatures fail at this step, because their performance in discovery was real only for the samples used to find them.

History has made the point expensively. In 2002 Petricoin and colleagues reported in The Lancet that a pattern of proteins in serum, read by mass spectrometry, detected ovarian cancer with near-perfect accuracy. They found the pattern in serum from 50 women with ovarian cancer and 50 without, and on 116 masked samples it identified all 50 cancers and 63 of the 66 other women. Six of the paper's 11 authors came from the US National Cancer Institute and the FDA, and two were then at Correlogic Systems, the company that planned to offer the test and prepared it for commercial launch within two years. In 2004, however, the FDA said the test was a medical device subject to premarket review. Independent reanalyses, one of them by Baggerly, raised concerns that the separating patterns reflected differences in how and when the samples had been processed, and the test never reached the market as planned.

The Duke University trials that open Part VII ran in the late 2000s on gene-expression signatures developed at Duke. Findings of research misconduct followed: in 2015 the US Office of Research Integrity found that a former Duke researcher had engaged in misconduct in research funded by the National Institutes of Health. The case led directly to a 2012 report of the Institute of Medicine, requested by the National Cancer Institute, on the evidence standards that translational omics should meet.

The evidence ladder

A widely used framework for genetic and genomic tests asks three questions in order. Analytical validity asks whether the test measures the analyte, the substance it is meant to measure, accurately and reproducibly, in the specimen type it will receive, across operators, instruments, lots and days. Clinical validity asks whether the result corresponds to the condition in the intended population, with known sensitivity, the share of people with the condition whom the test calls positive, and known specificity, the share without it whom the test calls negative (Chapter 13). Clinical utility asks whether acting on the result improves outcomes for patients, compared with not testing. Ethical, legal and social considerations run alongside all three. The framework comes from the US Centers for Disease Control and Prevention, which named it ACCE after its four parts.

47 — From signature to test

ReadoutFocal detail

Each rung needs a different kind of study, and the rungs get harder going up. Analytical validation is laboratory work, and Chapter 19 shows how it is done, one measure at a time. Clinical validation needs well-characterized samples from the right people, while clinical utility often needs a prospective trial in which testing changes what is done and outcomes are followed. The expression signatures of Chapter 21 entered guidelines on the strength of studies of archived trial samples. Two randomized trials in thousands of women later showed that many women whose tumors scored at low or intermediate genomic risk could forgo chemotherapy, although some younger women with intermediate scores gained from it (Chapter 21).

The arithmetic of screening

Screening healthy people raises the bar further, because the disease is rare in the population being tested, and rarity changes what a positive result means. The number that matters to a person who tests positive is the positive predictive value, the share of positive results that are true.

These numbers are close to those reported for a real test. NHS-Galleri, a randomized, double-blind trial in eight regions of England, enrolled about 142,000 adults aged 50 to 77 who had not had cancer diagnosed or treated in the previous three years and were not being assessed for possible cancer. Half of them, assigned at random, had their blood tested at up to three yearly visits with a test designed to detect many cancers from one draw. The blood of the other half was stored, and both groups received usual care. The Cancer Research UK Cancer Prevention Trials Unit at Queen Mary University of London ran the trial with King's College London and Guy's and St Thomas' NHS Foundation Trust, and GRAIL, the test's maker, funded it. The investigators reported the results in two papers in September 2026, and three of their 13 authors were affiliated with GRAIL. The paper in Nature Medicine gives a positive predictive value of 52.0 % across the three screening rounds and a specificity of 99.50 % to 99.60 % in each round.

In The New England Journal of Medicine, Sasieni and colleagues reported that the trial's primary endpoint, the incidence of stage III or IV cancer of 12 prespecified types, did not differ significantly between the groups after three screening rounds. The screened group's rate divided by the control group's, the incidence rate ratio, was 1.03, with a p-value of 0.63. For stage IV cancer of those types, the key secondary endpoint, the ratio was 0.86, so the screened group's rate was about 14 % lower, but the 95 % confidence interval of the ratio ran from 0.74 to 1.00, and 1.00 is the value that means no difference. The journal's summary of the paper gives no p-value for that result and does not call it significant. GRAIL had presented it in May 2026 as nominally significant, meaning that it passed the conventional threshold of 0.05 before any allowance was made for the number of endpoints tested (Chapter 18).

On 23 September 2026 an FDA advisory panel voted on the test. A panel of this kind is a committee of outside experts whose vote advises the agency without binding it. As of October 2026 the agency's decision is pending, and the opening of Part IX gives the panel's votes (Chapter 25).

By contrast, a blood test for colorectal cancer that the FDA approved in 2024 targets one cancer. The FDA's summary of that approval reports results from 7,861 people in a study run by the maker. The test detected colorectal cancer 83 % of the time and gave a false positive to about 1 in 10 people without cancer or an advanced precancerous growth, and it is approved for average-risk adults aged 45 and older. Those figures describe the version approved in 2024, and in July 2026 the FDA approved a version that reads methylation alone (Chapter 25).

Routes to market differ by jurisdiction

A test reaches patients by different routes in the United States, the European Union and India, and the regulators differ in how much of the evidence ladder they review before it does. Part IX sets out the three as of October 2026. Chapter 25 covers the United States, where the FDA clears, grants or approves a test sold as a device, while a laboratory-developed test, one that a laboratory designs and runs itself, falls under the federal rules for clinical laboratories known as CLIA. Chapter 26 covers the European Union's In Vitro Diagnostic Regulation, its risk classes and the evidence that a notified body, the organization designated to assess a device, asks to see. Chapter 27 covers India's Medical Devices Rules and the licensing of tests by CDSCO, the national regulator, and by the state licensing authorities.

The pipeline is part of the test

In an omics test, the analysis software does much of the measuring: it calls the variants, normalizes the counts, applies the model and produces the result. Regulators treat the software accordingly. The FDA's 2018 guidance on sequencing-based tests for germline disease treats the bioinformatics pipeline as part of the test, to be documented, versioned and validated end to end and revalidated after changes. A signature defined by a model must also be locked before validation.

If the model is meant to change after release, US law since 2022 allows a predetermined change control plan, which sets out in advance what may change and how each change will be verified. The FDA finalized its guidance on such plans for software functions that use artificial intelligence in December 2024. India's software guidance of July 2026 asks that changes to software be documented as part of risk management, and it states that an Algorithm Change Protocol "may be devised, wherever applicable" (Chapter 27). Such a protocol can include a plan for retraining the algorithm, and the guidance makes it an option rather than a requirement.

+ Part VIII · The field in 2026

Owners, products and the next decade.

On 30 January 2026 Illumina completed its purchase of SomaLogic from Standard BioTools, and with the sale the aptamer platform of Chapter 9, whose binders are short strands of chemically modified DNA, passed to a company that sells sequencers. Thermo Fisher Scientific, whose Orbitrap mass spectrometer Chapter 3 describes, had completed its purchase of Olink, the maker of the proximity extension assay, on 10 July 2024. By 2026, therefore, both of the large affinity platforms that Chapter 9 compares belonged to makers of other kinds of instrument: a maker of sequencers and a maker of mass spectrometers. Illumina paid $350 million in cash for SomaLogic at closing, with up to $75 million more in milestones and royalties tied to performance. Six months later, in July 2026, Standard BioTools stated that Illumina had agreed to pay it about $30 million to buy out the performance payment for 2026 and the future royalties. Owners, product names and regulatory records in this field change within months, so this part gives every status as of October 2026. Chapter 23 sets out who builds and sells the instruments, kits and tests of each layer and which of them hold a regulator's decision, and Chapter 24 separates the changes likely in the next decade from the limits likely to remain.

+ The questionWho builds and sells the instruments, kits and tests of each omics layer in October 2026, and which of them has a regulator examined?

At least nine changes of owner since May 2024

At least nine omics businesses changed owner between May 2024 and July 2026, SomaLogic among them, and in October 2026 two more sales were waiting to close. Several of these businesses make products that earlier chapters describe, often without naming the maker: the proximity extension and aptamer panels of Chapter 9, the flow cytometers of Chapter 11, the imaging panels and split-pool kits of Chapter 12 and the EV-based prostate test of Chapter 14. The table lists each change with two dates kept apart, the day the agreement or the purchase was announced and the day the purchase was completed. A dash marks an announcement date that the buyer's release or filing on the completion does not give, and the dates in one row come from a news report.

Some businesses and products took new names through these sales. BD's biosciences business has traded as Waters Biosciences since 9 February 2026, and Illumina's release on the SomaLogic purchase states that Illumina Protein Prep is now called Illumina SomaSeq Discovery. A sale can also leave the seller with rights: when Bio-Techne announced the sale of Exosome Diagnostics on 5 August 2025, it stated that it would retain access to that company's vesicle technology for developing kits. Five of the completed purchases joined a business to a buyer that sells another kind of instrument. A sequencer maker bought an aptamer platform for proteins, three makers of mass spectrometers bought an affinity platform, a spatial business and a cytometry business, and a maker of immunoassays bought a company that images proteins in tissue.

48 — Five purchases that crossed techniques, 2024 to 2026

Focal detail

The sections that follow take the layers in the order of the book and keep three kinds of statement apart. Any specification, price, count or claim to be first that a company makes stays labeled as that company's, unless a regulator's record supports it. A regulator's decision is given with its date and the regulator's verb, which Chapter 21 introduced: the FDA clears a premarket notification, or 510(k), grants a De Novo request, and approves an application for premarket approval, or PMA. A product in the laboratory research phase, which its maker does not offer for diagnosis, is labeled research use only, with the statement that the FDA's labeling regulation sets for such products: "For Research Use Only. Not for use in diagnostic procedures." The sections below give that label only where a maker states it. Chapter 25 gives the regulation, and every status in this chapter is as of October 2026.

Sequencers, and the kits that read marks on DNA

Illumina's NovaSeq X Series is one of the production-scale short-read sequencers whose output Chapter 5 gives in terabases. Illumina specifies about 8 to 10.5 terabases from its 25B flow cell in a run that reads 150 bases from each end of every fragment, with at least 85 % of the bases above Q30, the quality score of Chapter 4. The company states a $200 genome at list price on that flow cell, without stating the coverage or what the figure includes. On 19 November 2013 the FDA granted De Novo requests for Illumina's MiSeqDx instrument and a universal reagent kit, and it cleared two cystic fibrosis assays that run on them.

Other makers launched or announced short-read instruments in 2025 and 2026, each with its own figures. Element Biosciences announced its VITARI system on 19 February 2026, with up to 10 billion reads per run and a system price of $689,000, and as of October 2026 the system was on pre-order for shipping in the second half of the year. The company calls VITARI the first high-throughput benchtop system to deliver a high-quality whole genome at $100, and its release does not say what the $100 covers. Ultima Genomics put its UG200 Series on sale in February 2026, at US list prices of $1,250,000 for the UG200 Ultra and from $850,000 for the standard UG200, with shipments from the second quarter. On 20 October 2026 Ultima launched its Solaris 2.0 Max consumables for the UG200 Series and for the earlier UG100, and the company states that they enable a $60 whole human genome at 30× coverage. Its release does not say what the $60 covers. Roche launched AXELIOS 1, the sequencing-by-expansion system of Chapter 5, on 29 June 2026 at a US system price of $750,000. Roche states a mean raw read accuracy above 99.3 % and describes two modes. In the duplex mode the two strands of each molecule are linked during library preparation, so that their reads can be combined for accuracy. The simplex mode is built for throughput, and for it Roche states reads of up to 1,500 bases under suitable sample and library conditions. Roche labels the system for research use only and publishes no cost per genome.

Complete Genomics unveiled the DNBSEQ-T1+ and DNBSEQ-E25 Flash sequencers in February 2025, and on 23 February 2026 it agreed to be acquired by Swiss Rockets, a company based in Basel. Complete Genomics is the US subsidiary of MGI Tech. The agreement required the approval of MGI Tech's shareholders, and on 19 August 2026 Complete Genomics reported that they had given it. The purchase still needed clearance from the Committee on Foreign Investment in the United States, which reviews foreign investment in US businesses, and as of October 2026 the sale was pending.

A US law will limit federal business with some sequencer makers. The BIOSECURE Act became law on 18 December 2025 as section 851 of the defense authorization act for 2026, a section titled "Prohibition on contracting with certain biotechnology providers". On 10 December 2025, before the act was signed, the law firm Covington & Burling published a summary of the act's final text. Covington describes the prohibition as applying to federal agencies, and to their contractors and subcontractors in the work they do on federal contracts. By Covington's reading, the act does not bar a contractor from all business with a biotechnology company of concern, the act's term for a provider it covers. In a summary of January 2026 the law firm Ropes & Gray notes that the act names none of BGI, MGI or Complete Genomics, and that a company can come under it through a Defense Department list of Chinese military companies. The list that the department published in June 2026 names BGI Group, with Complete Genomics among its subsidiaries, and MGI Tech, while the earlier list that Ropes & Gray described did not name Complete Genomics. The government's own list of biotechnology companies of concern is due by 18 December 2026, and the act's prohibitions take effect only after the federal acquisition rules are revised.

PacBio and Oxford Nanopore make the long-read sequencers of Chapter 5. PacBio states that its SPRQ-Nx chemistry for the Revio system has shipped worldwide since 26 May 2026, at a list price of about $345 for a human genome at 20× coverage, and under $300 for a laboratory sequencing 5,000 genomes a year with discounts. The company states that these figures cover sequencing reagents only, leaving out the instrument, library preparation and labor. It gives its HiFi reads as typically 15,000 to 20,000 bases long and 99.9 % accurate, with methylated cytosines read from the kinetics of the polymerase. PacBio is also the maker in Chapter 24 that announced, on 4 November 2025, a Class III registration in China for its Sequel II CNDx system, and the "world's first regulatory clearance of a clinical-grade long-read sequencer" is PacBio's own description of it.

Oxford Nanopore states a raw read accuracy of 99.75 % with its latest basecalling software, and reads longer than 100,000 bases. In January 2026 it announced that its GridION Dx instrument carries the CE and UKCA markings, under which a device is sold in the European Union and in Great Britain, and called it the company's first diagnostic device registered in the UK and Europe. The announcement names no device class and no notified body, the independent assessor of Chapter 26, so it does not show whether the company declared the marking itself or had it certified. The first application the company names is the characterization of infectious disease, offered through specified partners, and it states that its tuberculosis assay, AmPORE-TB, remains for research use only.

Methylation is read either after a conversion step or directly from the native molecule, and makers sell both routes (Chapter 6). Illumina specifies that its Infinium MethylationEPIC v2.0 array targets about 930,000 methylation sites from 250 ng of DNA, with eight samples on each array. New England Biolabs sells the enzymatic conversion kit of Chapter 6 as the NEBNext Enzymatic Methyl-seq v2 Kit, which it specifies for inputs of 0.1 to 200 ng. The paper that Chapter 6 reads on the method was written by the company's staff and funded by the company. On 15 October 2025 Illumina released a "5-base solution", which it states detects genetic variants and DNA methylation from one sample, the kind of method that Chapter 24 dates to 2025. Among the native readers, Oxford Nanopore gives 99.2 % accuracy for its methylation calls on synthetic DNA whose marks are known. RNA sequencing runs on the same instruments, with library kits from many reagent makers (Chapter 7).

Mass spectrometers, binding panels and metabolite services

Makers of mass spectrometers launch new instruments at the yearly conference of the American Society for Mass Spectrometry, and trade publications report the makers' figures. At the 2025 meeting in Baltimore, Thermo Fisher Scientific launched the Orbitrap Astral Zoom, which it describes as scanning up to 270 times a second. Bruker launched the timsUltra AIP, the timsOmni and the timsMetabo, which it describes as scanning up to 300 times a second, and SCIEX launched the ZenoTOF 8600, which it says has ten times the sensitivity of its earlier ZenoTOF instruments.

At the 2026 meeting in San Diego, from 31 May to 4 June, Thermo Fisher introduced the Orbitrap Tribrid Apex, which a trade report describes as collecting data four times faster. Waters introduced the Xevo MRT P10 and states that it gives up to 20 times more sensitivity in fragment scans. Bruker introduced the timsMRMS, and Shimadzu introduced software for untargeted metabolomics and lipidomics. None of these reports gives samples per day or a count of proteins identified, and the number of proteins an instrument finds in plasma depends on the sample and the method as well as on the instrument (Chapter 8).

Other companies sell the steps before the spectrometer. Seer sells one form of the nanoparticle enrichment of Chapter 8. It states that proteins from a sample bind to its engineered nanoparticles as a layer called a protein corona, and that the binding compresses the sample's dynamic range. Seer launched its Proteograph ONE assay and SP200 automation instrument at the 2025 meeting, and a trade report gives Seer's claim that the instrument processes more than 1,000 samples a week, with an automated run of under 5 hours. Evosep, which makes separation systems for proteomics, launched its Eno platform the same year and states that its standardized methods run between 30 and 500 samples a day. Throughputs of this kind lie behind Chapter 24's "hundreds of samples a day", and they are the makers' own figures.

Olink makes the proximity extension assay of Chapter 9, SomaLogic makes the aptamer platform, and UK Biobank's expansion measures up to 5,400 proteins on Olink's Explore HT platform. Alamar Biosciences makes the method of Chapter 9 that is related to the proximity extension assay and adds purification steps, and it sells NULISAseq panels whose names give their protein counts: Neuro 220, CNS Disease 120, Immune 340, Inflammation 250 and a mouse panel of 120. Alamar's site marks its products for research use only. In July 2026 the company announced that it would profile about 21,000 plasma samples from 10,000 participants in dementia studies with its Neuro 220 panel, for a consortium funded by the US National Institute on Aging, and that the data would be public.

Quantum-Si and Nautilus Biotechnology are the two companies of Chapter 9 that read proteins one molecule at a time. Quantum-Si sells the Platinum protein sequencer, and Chapter 9 reports its development results for the next instrument, Proteus, which the company expects to launch in the second quarter of 2027. In its second-quarter results of 28 July 2026, Nautilus stated that its Voyager platform was not yet commercially available. It expected to take pre-orders in early 2027 and to ship instruments in mid-2027 with kits for measuring proteoforms, and it said that the timeline for its assay of the broad proteome would be reported later.

Metabolon and biocrates sell metabolomics of the kinds Chapter 10 describes, as services and as kits. Metabolon runs its services on liquid chromatography with tandem mass spectrometry, and it states that its untargeted panel covers up to 70 major biochemical pathways and that its targeted panels give absolute amounts for more than 1,000 metabolites and lipids. The company gives its reference library as over 5,400 metabolites and calls it the world's largest, and it also sells a profiling kit, Verus. The Innsbruck company biocrates sells targeted kits such as the MxP Quant 500 and the AbsoluteIDQ p180, as well as services, and its logo carries the words "Biognosys group" without the site explaining the link. Both companies state that their products and services are for research use only.

Cytometers, single cells and tissue sections

The flow cytometers of Chapter 11 come from several makers. The spectral instrument used for Chapter 11's 50-color panel came from BD Biosciences, which is now Waters Biosciences. Waters gained that business through a combination in which BD's biosciences and diagnostic businesses were separated from BD and merged with a Waters subsidiary, and the diagnostic business joined Waters Advanced Diagnostics. Cytek Biosciences is the manufacturer that Chapter 11 says placed a 60-color instrument in early access in 2026. It announced the Borealis spectral cytometer, with 7 lasers and 120 detectors, on 2 June 2026, for a debut at the CYTO 2026 meeting.

Mass cytometry, the metal-tag method of Chapter 11, belongs to Standard BioTools, which sells the CyTOF and Hyperion instruments. On 30 September 2026 Standard BioTools ended an agreement of July 2026 to sell that business to Multiplex Bio, and agreed instead to sell it to Element Biosystems, an acquisition company that GMT Venture Partners formed. Element Biosystems is a different company from the sequencer maker Element Biosciences. Standard BioTools expects the sale to close by the end of 2026, together with its own merger with Treeline Biosciences and subject to its stockholders' approval, and until then it remains the seller of both instruments.

10x Genomics makes the Chromium droplet systems of Chapter 12, and it is the maker that stated in October 2024 that its GEM-X Flex chemistry profiles up to 2.5 million cells per run, at as little as one cent a cell. On 9 June 2026 it announced the purchase of Proteintech Genomics, a division of Proteintech Group, without disclosing terms, and it calls that division's antibody panel the largest of its kind for single-cell protein. Qiagen now owns Parse Biosciences, the maker of the split-pool Evercode kits of Chapter 12, which need no instrument, and Waters Biosciences now sells the single-cell multiomics products that BD sold.

Bruker has sold NanoString's spatial and gene-expression instruments since it bought that business's assets in 2024, through a unit now called Bruker Spatial Biology. The unit's products are the CosMx imager, the GeoMx profiler, the nCounter system and CellScape, and its site labels them for research use only. The imaging panels of Chapter 12 that reached about 6,000 genes in 2024 and the whole transcriptome in 2025 are CosMx panels. Chapter 12's 5,000-gene panel is 10x Genomics' Xenium Prime 5K, launched in May 2024 for 10x's own imaging instrument. 10x introduced Atera on 18 April 2026 and describes it as a platform for whole-transcriptome spatial analysis with single-cell sensitivity. On 29 September 2026 the company announced its first commercial shipments of Atera, and neither that announcement nor the one at launch gives a price or a research-use statement.

In 2025 Quanterix, which sells its own Simoa immunoassays, bought Akoya Biosciences, the maker of the PhenoCycler and PhenoImager systems for spatial proteomics. Vizgen, the private maker of the MERSCOPE imager, launched new panels and segmentation tools in April 2026, and its news page lists a report that it raised $48 million in January 2026. Two sequencer makers have spatial plans as well. Complete Genomics agreed in March 2025 to supply Stereo-seq spatial transcriptomics to the Human Cell Atlas, and Element Biosciences lists direct spatial sequencing in tissue on its AVITI24 system as a plan rather than a launched product.

Blood and tissue tests that hold FDA decisions

The colorectal blood tests of Chapters 14, 21 and 22 that the FDA approved in 2024 and 2026 come from two companies. Guardant Health's Shield was approved on 26 July 2024, and its change to a methylation-only model on 22 July 2026 (Chapter 25). Freenome's SimpleScreen CRC was approved on 24 July 2026, four days after Freenome became a company listed on Nasdaq through a merger with Perceptive Capital Solutions, a listed company formed to acquire a private one. Freenome's filing on the listing includes a collaboration and license agreement with Exact Sciences dated 3 August 2025, without stating its terms or which company markets the test. Exact Sciences, part of Abbott since March 2026, holds the FDA's approvals of two earlier colorectal screening tests that analyze stool: Cologuard, approved on 11 August 2014 for adults aged 50 and older, and Cologuard Plus, approved on 3 October 2024 for adults aged 45 and older at average risk.

Abbott's announcement of the purchase also names Oncodetect, which it describes as a tumor-informed test for molecular residual disease, and Cancerguard, a blood test for detecting many cancers, without giving their regulatory status. The FDA's PMA records under Exact Sciences' name include neither of the two. GRAIL's Galleri, the multi-cancer test of Chapters 22 and 25, was awaiting the FDA's decision on its PMA as of October 2026.

The profiling tests of Chapter 21 come from three makers. Foundation Medicine holds approvals for FoundationOne CDx, a tissue test of 324 genes approved on 30 November 2017, and for FoundationOne Liquid CDx, a plasma test of 311 genes approved on 26 August 2020. Guardant Health holds approvals for Guardant360 CDx, a plasma test approved on 7 August 2020, and for Guardant360 Liquid CDx, a plasma panel of 741 genes approved on 19 May 2026. Illumina holds the approval of 21 August 2024 for TruSight Oncology Comprehensive, the distributable kit of Chapter 21. Natera's Signatera CDx, approved on 15 May 2026, is the residual-disease companion diagnostic of Chapter 21. Its PMA is the only one in the FDA's database under Natera's name, so Signatera outside its approved use, and Natera's other tests such as its prenatal screen, carry no FDA approval on that record.

Four of the FDA's decisions in this chapter, all on new tests, came within about ten weeks in 2026: the clearance of an Indian company's profiling test on 12 May, described below, the two approvals of 15 and 19 May, and the approval of SimpleScreen CRC on 24 July. A news service reported that CareDx completed its purchase of Naveris in July 2026. Naveris's NavDx test detects tumor-derived circulating DNA to monitor cancer, and the report gives no regulatory status for it. The plate below gathers the FDA's decisions on the products in this chapter.

49 — FDA decisions for the products in this chapter, 2013 to 2026

ReadoutFocal detail

No test built on extracellular vesicles appears among these decisions. Since September 2025 mdxhealth has owned the ExoDx Prostate test, together with a CLIA-certified laboratory. The test is the EV-based one whose 2019 designation Chapters 14 and 25 describe, and mdxhealth now sells it as Exo mdx. The company states that the test combines the expression of three genes, PCA3, SPDEF and ERG, measured in RNA from EVs in urine collected without a rectal examination. Neither mdxhealth's announcement nor Bio-Techne's states the test's FDA status, and the search reported in Chapter 14 found no FDA clearance or approval of an EV-based diagnostic as of October 2026.

India: service laboratories, one clearance and a national data set

The Indian companies in this chapter run laboratories that sequence samples and report results, and the figures that follow are their own. MedGenome, which runs laboratories in Bangalore and in Foster City, California, raised $47.5 million in July 2025 in a funding round led jointly by Maj Invest and Novo Holdings. The announcement, from Novo Holdings, carries MedGenome's description of itself: more than 1,300 genetic tests, more than 500,000 genomes and exomes sequenced over the last decade, and a Bangalore laboratory that MedGenome calls the largest genetic testing laboratory in India accredited by CAP, the US accreditation body of Chapter 19. Strand Life Sciences, also in Bangalore, sells clinical sequencing reports for inherited and tumor variants, software for omics and biomarker services for drug companies. It states that its sequencing laboratory is accredited by CAP.

4baseCare, a third Bangalore company, profiles tumor tissue and circulating tumor DNA, and states that it has mapped more than 30,000 genomic profiles. Mapmygenome sells consumer and clinical genomic tests, among them a whole-genome sequence at 89,000 rupees without the raw data. Its site describes its laboratory as accredited by NABL and certified to the 2012 edition of ISO 15189, an edition that ISO has superseded and that Chapter 27 shows NABL no longer accredits against.

An Indian company received an FDA clearance for an omics test in 2026. On 12 May 2026 the FDA cleared CellDx-Tissue, a sequencing-based tumor profiling test from Datar Cancer Genetics of Nashik, by a 510(k) that went through third-party review, in which an organization accredited by the FDA reviews the submission before the agency decides. It is the clearance that Chapter 25 cites under the classification "Next Generation Sequencing Based Tumor Profiling Test". Datar describes itself as "India's only US FDA-cleared comprehensive genomic profiling assay provider", and that claim is the company's own. The clearance covers CellDx-Tissue alone. The company also sells two blood tests, one for detecting tumors in suspected cases and one for screening people without symptoms, and no FDA record for them was found under the company's name. Datar states that its laboratory in Nashik is CLIA-certified and CAP-accredited, and that a second laboratory, in Guildford, England, is registered with the Care Quality Commission, England's regulator of health and social care.

The national genome data set of Chapter 4 comes from GenomeIndia, a project funded by the Department of Biotechnology, and holds about 10,000 whole genomes. A search for this guide in October 2026 found no Indian maker of a sequencer, mass spectrometer, cytometer or omics kit, and returned procurement tenders, distributors and market reports instead. One search cannot show that no such maker exists, and it left out makers of PCR and immunoassay kits, whose techniques lie outside this chapter. A test kit made or imported for sale in India needs a license under the Medical Devices Rules of 2017, and Chapter 27 sets out who issues it and what research use only means there.

+ The questionWhat will change in omics over the next decade, and what is still unsolved?

A rule for reading the future

On 8 October 2026 a developer of single-molecule protein sequencing announced that it had detected 19 of the 20 amino acids in its development program, three weeks after announcing 18 with a developmental kit (Chapter 9). Both announcements were the company's own statements about work in development. Claims about the future of omics arrive faster than evidence, and they look alike: a press release announcing a shipment reads much like one announcing a plan.

The most useful habit this chapter can offer is therefore a sorting rule. For any claim, the first question is which of four columns it belongs in: in routine use; available and in early use; demonstrated in research; or announced, projected or under review. The claim is then read with the evidence that column carries, and the plate below applies the rule to the developments discussed in this guide, as of October 2026.

50 — Demonstrated, available, announced

DNARNAProteinMetaboliteModificationFocal detail

Sequencing: cheaper, longer, and read with its marks

The cost of reading DNA is likely to keep falling, though the easy gains have probably been taken (Chapter 5 traces the fall so far). The remaining costs are likely to lie increasingly in sample handling, analysis, storage and interpretation rather than in reagents. The larger changes are in what a read contains. Long reads are entering the clinic: in November 2025 a sequencer maker announced that its long-read system had received a Class III registration in China, the country's class for the highest-risk devices. The registration covers testing for thalassemia, an inherited blood disorder, and the maker called that registration the first regulatory clearance of a clinical long-read sequencer.

Methods that report methylation and sequence from the same library, launched in 2025, have begun to make the epigenome a by-product of reading the genome (Chapter 6). Pangenome references are beginning to supplement single references (Chapter 4). Population programs now include more than ten thousand long-read genomes alongside hundreds of thousands of short-read ones.

Proteins: population scale, and single molecules

Protein measurement is where the next decade will change most, because it starts furthest behind. Population proteomics will move from tens of thousands to hundreds of thousands of people, with the full data of the UK Biobank expansion expected in 2027 (Chapter 9). Mass spectrometers keep gaining speed, and the makers of the separation and sample-preparation systems that feed them report throughputs of hundreds of samples a day (Chapter 23). Single-molecule protein sequencing is the development to watch most carefully, because it attacks the limit set out in Chapter 2. It has moved from academic demonstration to shipping instruments with partial amino-acid coverage, and the 2026 announcements that open this chapter are development results. Reading a plasma proteome molecule by molecule, at depth and at cost, remains a goal rather than a result.

Cells at atlas scale, and the models trained on them

Single-cell experiments have reached the scale of atlases. In October 2024 one maker stated that its droplet system can profile up to 2.5 million cells in a run, and a company's 2025 preprint described a dataset of about 100 million cells from 50 cancer cell lines exposed to drugs (Chapter 12). Much of that scale is being spent on perturbation atlases, which measure systematically how cells respond when each gene is switched off or each drug is applied. Spatial methods are converging on whole-transcriptome coverage at cellular resolution, and on measuring RNA, protein and chromatin in the same section. The limiting factors are shifting from chemistry to segmentation, the drawing of each cell's boundary in an image, and to data volume and cost per tissue area.

Machine learning has already changed one part of molecular biology decisively: protein structure prediction, which in 2024 was extended to complexes of proteins with nucleic acids and small molecules. One such model, AlphaFold 3, was described that year by Abramson and colleagues, who worked at Google DeepMind and Isomorphic Labs, the two companies that built it. Models that read long stretches of DNA sequence and predict regulatory activity, or score the likely effect of a variant, were published in 2025 and 2026. Google DeepMind described one of them, AlphaGenome, in a preprint in June 2025 and in Nature in January 2026. It reads up to a million bases at once and predicts about 5,900 human genomic signals, such as gene expression and splicing, and its authors report that it matched or beat earlier models in 25 of 26 tests of variant effects. Another, Evo 2, built by the Arc Institute with Stanford University and NVIDIA among others, was trained on 9.3 trillion DNA bases, and its authors report that it predicts the effects of variants without being trained for that task.

A more ambitious goal is a "virtual cell", a model that predicts how any cell will respond to any perturbation, and it is announced rather than demonstrated. Independent benchmarks published in 2025 found that several large single-cell foundation models, models trained on many single-cell datasets at once, did not outperform simple linear baselines at predicting the effects of genetic perturbations. In one of those benchmarks, Ahlmann-Eltze, Huber and Anders tested five such models and two other deep-learning models on how the transcriptome changes after one or two genes are perturbed, and none beat the deliberately simple baselines. The three were academic researchers at EMBL and Heidelberg University when they published the benchmark.

The Arc Institute ran its first Virtual Cell Challenge as an open competition on the same task, with sponsorship from companies that include makers of single-cell and sequencing instruments. Teams predicted how human embryonic stem cells respond when single genes are knocked out, and when the winners were announced in December 2025, the winning models did not beat naive baselines on every metric either.

The 2025 results suggest that the binding constraint is data designed to answer the question rather than model size, which is why the perturbation atlases above matter. In October 2026 Shift Bioscience, a company that builds virtual-cell models, announced a study published with collaborators in Nature Biotechnology. The company stated that deep-learning models that predict the effects of perturbations can beat uninformative baselines once the metrics are calibrated, although a simple additive baseline remained hard to beat on the dataset with the fewest gene combinations in its training data.

51 — What one challenge measured

ReadoutFocal detail

Liquid biopsy and vesicles

A blood test for many cancers at once is under review at the FDA, and its evidence will set the terms for the field. In May 2026 its maker presented the results of a randomized trial of the test among about 142,000 adults in England, and in September 2026 the trial's investigators published them in The New England Journal of Medicine. The trial missed its primary endpoint, a lower incidence of stage III or IV cancer. The rate of stage IV diagnoses, its key secondary endpoint, was 14 % lower in the screened group, but the confidence interval around that result reached the value that means no difference (Chapter 22). The maker funded the trial, and three of the paper's 13 authors were affiliated with the maker. As of October 2026 the FDA's decision on the test is pending (Chapter 25).

EV-based diagnostics are advancing more slowly, and the obstacle is largely metrological, a problem of measurement science, as Chapter 14 argued: calibrated detection limits, reference materials and reporting standards. Until all three are in place, counts of vesicles from different laboratories cannot be compared, whatever the biomarker.

What still needs to be solved

Most of the limits described in this guide are physical or statistical, rather than engineering problems waiting for the next platform, and naming them is the most durable part of any forecast.

• The protein dynamic range. Protein concentrations in plasma span about ten orders of magnitude, and proteins can be neither copied nor read by pairing (Chapter 2). Enrichment, affinity and single-molecule methods each move the limit, but none has removed it.
• Metabolite identification. One reanalysis of untargeted datasets from blood estimated that more than half of the compounds detected with high confidence are still unknown (Chapter 10).
• Reproducibility across laboratories. Batch effects, platform differences and unreported processing choices still stop many results from transferring between sites (Chapter 18).
• Detection limits for particles. Vesicle and nanoparticle measurements need calibrated, stated detection limits before their counts can be compared (Chapter 14).
• Diversity of reference data. Variant interpretation remains weakest for populations underrepresented in reference cohorts, including many in South Asia (Chapter 4).
• From validity to utility. Many signatures have analytical and clinical validity, and far fewer have evidence that acting on them improves outcomes (Chapter 22).
• Interpretation and privacy. A 2015 review expected the cost of interpreting variants to fall more slowly than the cost of generating the sequence, so that the two would become comparable. In two hospital costing studies of genome sequencing, one from the United Kingdom published in 2019 and one from Canada published in 2020, sequencing reagents and other supplies made up the largest share of the cost, and neither study costed genetic counseling. Reagent prices have fallen since (Chapter 5). Genomic data also raise privacy questions that outlast any single study.

+ Part IX · Quality and regulation, as of October 2026

What each regulator reviews, and when.

On 23 September 2026, from nine in the morning until six in the evening, an FDA advisory panel weighed a blood test that reads methylation patterns in cell-free DNA to screen adults aged 50 and older for many kinds of cancer. The test was GRAIL's Galleri, and the application before the panel was for premarket approval, the route the FDA reserves for devices in its highest risk class. A panel advises the FDA without binding it. The FDA's decision will close its review of the test's safety and effectiveness, the review that premarket approval requires, and as of October 2026 the agency had not decided. The FDA's summary of the meeting records three votes, each for the patients who meet the test's criteria. The panel voted 10 to 0 that the data show a reasonable assurance that the test is safe, 6 to 4 that they show a reasonable assurance that it is effective, and 7 to 2, with one abstention, that its benefits outweigh its risks. GRAIL reported the same counts that day. Chapters 19 and 22 set out two kinds of evidence for a test: a quality system that turns a claim into traced requirements and validated measurements, and a ladder of evidence that climbs from analytical validity through clinical validity to clinical utility. The three chapters of this part set each jurisdiction's rules against those two chapters, with their dates. Chapter 25 covers the United States, where a test reaches patients either as a device the FDA reviews or as a laboratory's own test. Chapter 26 covers the European Union's In Vitro Diagnostic Regulation, and Chapter 27 India's Medical Devices Rules. All three give regulatory status as of October 2026.

+ The questionDoes an omics test reach US patients as a device the FDA has reviewed or as a laboratory's own test, and what has each route checked before the first result is reported?

Two routes, after a court vacated the 2024 rule

On 31 March 2025 a federal court in the Eastern District of Texas vacated the FDA's rule on laboratory-developed tests, in a case brought by the American Clinical Laboratory Association. The FDA had published the rule on 6 May 2024. It had added a phrase to the regulation that defines in vitro diagnostic products, "including when the manufacturer of these products is a laboratory", so that a laboratory designing and running its own test became the manufacturer of such a product. On 19 September 2025 the FDA published a final rule that removed the phrase, "reverting to the text of the regulation as it existed prior to the effective date of the Rule". It called the step "ministerial in nature", because the court had already vacated the rule. As of October 2026 the FDA's page on laboratory-developed tests shows no later action, and a search for this guide found no action of 2026 by the FDA or by Congress.

52 — US rules and records for omics tests, May 2024 to October 2026

ReadoutFocal detail

The regulation left in place defines in vitro diagnostic products as "those reagents, instruments, and systems intended for use in the diagnosis of disease or other conditions". The definition turns on what a product is intended for, not on how it works, so a reagent kit, an instrument such as a sequencer, or a system that joins them with its analysis software can each be one. A product made and sold for diagnosis is a device, and its class decides what the FDA reviews before it is sold. A test that a laboratory designs and runs itself is what the FDA calls a laboratory-developed test, and it falls under CLIA, the law for US clinical laboratories (Chapter 17). As of October 2026, an omics test therefore reaches US patients by one of two routes, as a device or as a laboratory's own test.

The line between the two routes was argued long before the 2024 rule. A trade report of 2005 records that in 2004 the FDA had determined the serum proteomic test of Chapter 22 to be a medical device subject to premarket review, although its developer had hoped to offer it as a laboratory test under CLIA.

Products that are not yet sold for diagnosis carry one of two statements set out in the FDA's labeling regulation. A product in the laboratory research phase is labeled "For Research Use Only. Not for use in diagnostic procedures." A product being tested before full commercial marketing is labeled "For Investigational Use Only. The performance characteristics of this product have not been established." The second statement thus tells every user, on the label itself, that the product's performance has not yet been shown, and Chapter 27 returns to it.

Three classes, three routes, three verbs

The FDA's regulation on classification, 21 CFR 860.3, defines three classes by the controls each needs. "Class I means the class of devices that are subject only to the general controls" of the law. "Class II means the class of devices that is or eventually will be subject to special controls", the requirements the FDA writes for one type of device (Chapter 21). Class III is the class "for which premarket approval is or will be required". The FDA describes general controls as the basic provisions of the Medical Device Amendments of 28 May 1976, and they apply to devices in all three classes unless a device is exempted. They include the provisions on adulteration and misbranding, registration and listing, premarket notification, records and reports, and good manufacturing practice. The FDA's classification regulations for in vitro diagnostics sit in Parts 862, 864 and 866 of Title 21, and each names a type of device and the class it belongs to.

The class sets the route, and each route ends in a decision with its own verb. In a 510(k), or premarket notification, the maker compares its device with a predicate, a legally marketed device that is not subject to premarket approval, and when the FDA finds the two substantially equivalent, its decision is a clearance. A novel device without a predicate can be classified through a De Novo request, and when the FDA agrees, it grants the request. The FDA describes the route as "a pathway to classify novel medical devices" for which general controls, alone or with special controls, give reasonable assurance of safety and effectiveness. A class III device needs premarket approval, or PMA, the FDA's review of its safety and effectiveness, and when the review succeeds the FDA approves the maker's application. In each case the agency decides, and the company receives the decision.

All three routes carry omics tests. In 2007 the FDA granted a De Novo request for a 70-gene expression signature, and it has approved by PMA the blood tests for colorectal cancer and the companion diagnostics of Chapter 21. On 12 May 2026 it cleared by 510(k) a test classified as a "Next Generation Sequencing Based Tumor Profiling Test", under 21 CFR 866.6080 (Chapter 23). Each test's route follows from its claim, through the class that the FDA assigns to a device of its type. For some companion diagnostics that type is now defined by technique, which the next section describes.

For sequencing tests, two final guidances that the FDA issued in April 2018 set out the evidence it expects, and they correspond to the first two rungs of Chapter 22's ladder. The first covers the design, development and analytical validation of sequencing tests that aid the diagnosis of suspected germline disease, and Chapter 19 sets out its measures. The second covers the use of public databases of human genetic variants to support clinical validity, for genetic and genomic tests in general, and the FDA runs a voluntary process for recognizing such databases. Neither guidance had been withdrawn as of October 2026.

53 — Two US routes for an omics test, and what each checks

Focal detail

Companion diagnostics, reclassified one technique at a time

Companion diagnostics for cancer drugs have been class III devices in the United States, approved by PMA, and the FDA's guidance on in vitro companion diagnostics dates from 2014. The claims on a companion diagnostic's label can be extended one drug at a time, each through a supplement that the FDA approves. In September 2026 the FDA approved two supplements to the PMA of one blood-based panel, Guardant360 CDx, and each added a companion claim for a gene called ESR1. The first, on 4 September, paired the gene with the drug camizestrant, and the second, on 18 September, with imlunestrant given with abemaciclib.

The FDA has begun to move these tests down a class, one technique at a time, by orders that reclassify a type of device. On 17 August 2026 it published a final order that places test systems based on in situ hybridization, the probe-pairing method of Chapter 21, in class II with special controls when they are used with an approved cancer drug. The order is codified as 21 CFR 864.1890 and covers four product codes, the three-letter codes by which the FDA groups devices of one type. A policy blog reports that it took effect on 16 September 2026, and that the FDA had proposed it on 11 June 2025.

The parallel order for nucleic-acid tests, sequencing panels among them, is still a proposal. The FDA published it on 25 November 2025. It would move nucleic-acid-based test systems used with an approved cancer drug from class III to class II with special controls, "subject to premarket notification", which is the 510(k) route. No final order had been found as of October 2026, and a policy blog of September 2026 reported none. One of the four product codes the proposal covers, PQP, is the code under which the FDA approved two companion diagnostics by PMA in May 2026, a test for molecular residual disease on 15 May and a blood panel covering 741 genes on 19 May (Chapter 21). Those approvals came about six months after the proposal, and until an order is final, the route it proposes does not apply.

The FDA also grants some devices a Breakthrough Device designation before it decides on them. The FDA describes its Breakthrough Devices Program as a voluntary program for devices that provide more effective diagnosis or treatment of a life-threatening or irreversibly debilitating disease and that meet at least one further criterion, such as having no approved or cleared alternative. A maker whose device is designated can interact with FDA experts during development and can expect prioritized review of its later submissions, but the device must still meet the FDA's standards of safety and effectiveness. By 30 June 2026 the FDA had granted 1,320 designations and had issued 210 marketing authorizations for designated devices.

Two omics tests in this guide show that a designation and a decision on the test can be years apart. The company that then made the EV-based prostate test of Chapter 14 announced the test's designation in 2019, and a search for this guide found no EV-based test cleared, granted or approved by October 2026. A news report of September 2026 states that Galleri was designated in 2018. Its PMA was still awaiting a decision eight years later. A designation is therefore not a decision on the test: by designating a device, the FDA has not cleared, granted or approved it.

A laboratory's own test under CLIA

CMS, the federal agency that runs the CLIA program, defines three categories of test complexity. Its regulation lists waived tests, tests of moderate complexity, including a subcategory of microscopy performed by the provider, and tests of high complexity. The FDA's center for devices assigns the category, scoring a test system on seven criteria from 1 to 3 each, and a total of 12 or less makes the test one of moderate complexity. The FDA states that "Test systems that have not been CLIA categorized default to high complexity until they have been categorized." CMS states that laboratory-developed tests are considered tests of high complexity, and CLIA's regulation treats any test that is not on the lists of categorized tests as one of high complexity until it has been reviewed.

The category sets what the laboratory needs before it may run the test. A laboratory that runs tests of high complexity must hold a certificate for them and must have a laboratory director, a technical supervisor, a clinical consultant, general supervisors and testing personnel who meet the qualifications the regulation sets for that category. For tests of moderate complexity, by contrast, the regulation accepts testing personnel with a high school diploma and documented training. A laboratory that builds its own omics test therefore needs the certificate and the staff that tests of high complexity require.

What the laboratory must show is analytical. Before it reports any patient result from a test it built or modified, the laboratory establishes the performance specifications listed in Chapter 19, from accuracy and precision to the reference intervals. CMS reviews them at its routine survey every two years, after testing has started. For an unmodified test that the FDA has cleared or approved, the laboratory instead verifies that it obtains figures comparable to the maker's (Chapter 17). CMS states that its CLIA program does not address the clinical validity of any test (Chapter 19). For a laboratory's own test, the federal check reaches the first rung of Chapter 22's ladder, and it does so after use has begun.

The two routes carry the two quality systems of Chapter 19. The maker of a device, such as a kit or an instrument sold for diagnosis, works under the Quality Management System Regulation, which took effect on 2 February 2026 and incorporates ISO 13485:2016. On that day the FDA also stopped inspecting device makers with its older Quality System Inspection Technique and began to inspect them under an updated compliance program. A laboratory's own test rests instead on the laboratory's validation, checked by CMS, and on any accreditation the laboratory adds, such as the checklist requirements of CAP for sequencing tests (Chapter 19).

Galleri and Shield in the FDA's records

The FDA describes Galleri, the test before the panel at the opening of this part, as a "qualitative, next-generation sequencing (NGS)-based in vitro diagnostic test". Its bulletin of 21 September 2026 adds that the test is for prescription use and detects cancer-specific methylation patterns in cell-free DNA from whole blood, and it gives the intended use as "screening for the early detection of multiple types of cancer in adults aged 50 years or older". The meeting of the FDA's Molecular and Clinical Genetics Panel was public. Comments received in the meeting's public docket by 8 September went to the panel, and the docket closed on 16 September.

GRAIL states that it submitted the PMA on 29 January 2026. In August 2026 the trade publication 360Dx reported the company's account of the evidence in the application. By that account, the application drew on the first 25,000 participants of GRAIL's PATHFINDER 2 study, on the first year of NHS-Galleri, the randomized trial in England whose results Chapter 22 reads, and on an analysis that bridges those data to an updated version of the test. That evidence reaches past the analytical rung. NHS-Galleri enrolled about 142,000 adults and was built to show whether screening reduced the cancers diagnosed at late stages, which is a question of clinical utility. In September 2026 its investigators reported in The New England Journal of Medicine that the trial missed that primary endpoint (Chapter 22). GRAIL funded the trial, and three of the paper's 13 authors were affiliated with the company. After the vote GRAIL wrote that the FDA "is expected to make a final decision on the Galleri PMA in the coming months", and the 360Dx report of August had given the company's expectation of approval in the "first part of 2027". As of October 2026 no decision had been posted.

Shield, Guardant Health's blood test for colorectal cancer, shows how an approved test changes. The FDA received its PMA on 10 March 2023 and approved it on 26 July 2024 for adults aged 45 and older at average risk, with the statement that "Shield is not a replacement for diagnostic colonoscopy or for surveillance colonoscopy in high-risk individuals". On 14 November 2025 the FDA received a supplement, and on 22 July 2026 it approved "a change to the Shield EP1 classification model which streamlines Guardant Shield to a methylation-only workflow, removing fragmentomic and somatic components". The model that the 2024 approval had locked changed only through a new FDA decision. The decision was made on a 180-day supplement, the kind of supplement the FDA requires for a significant change in design, specification or software. The FDA reviews such a supplement in depth and approves it before the change is made. The performance figures in Chapter 22 belong to the version approved in 2024.

A smaller change took a lighter form. On 24 July 2026 the FDA approved the third blood test for colorectal cancer screening, which Chapter 21 describes. By 26 August 2026 the test's record also held a 30-day notice for a change in a supplier's reagent process. A 30-day notice is the route for changes to manufacturing methods, and the maker may put the change into effect 30 days after the FDA receives the notice, unless the FDA finds the notice inadequate. In both records the change went to the FDA before it was made, but only the new classification model needed the FDA's approval. Chapter 20 opens with the supplier's change to the third colorectal test and describes the review that a maker itself gives such a change.

+ The questionWhat does the European Union's regulation ask of an omics test before it reaches a patient, who checks the answer, and when may a hospital laboratory make its own?

Four classes, and the rule that places three kinds of omics test in class C

The European Union's In Vitro Diagnostic Regulation sorts every test into one of four classes of risk, from A to D, and three kinds of test that recur in this guide fall in the third. Under Rule 3 of the regulation's Annex VIII, which holds its classification rules, human genetic tests, companion diagnostics and tests for the screening, diagnosis or staging of cancer are class C, unless Rule 1 or Rule 2 places them in class D. The regulation is Regulation (EU) 2017/746, known as the IVDR, and Chapter 20 describes its rules for surveillance after sale. A sequencing panel for inherited variants, a test that selects a cancer drug and a blood test that screens for cancer therefore all fall in class C under Rule 3, whichever technique each one uses. Classes A and B hold tests of lower risk and class D those of the highest, and the class decides who must assess a test before it is placed on the market.

Chapter 21 introduced the term companion diagnostic, and the regulation defines it in its own words as "a device which is essential for the safe and effective use of a corresponding medicinal product". The United States and the EU reach the class of such a test from different starting points. The FDA classes a type of device by regulation and by order, and in 2026 it moved companion diagnostics based on in situ hybridization to class II by an order that names the technique (Chapter 25). The EU's rule names a purpose instead, so a companion diagnostic is class C whether it pairs a probe, sequences a panel or measures a protein.

Who assesses a test: the notified body

For every class but the lowest, and for sterile tests in the lowest, the IVDR requires an assessment by a body other than the test's maker. That body is a notified body, a conformity assessment body designated under the regulation. The IVDR defines conformity assessment as "the process demonstrating whether the requirements of this Regulation relating to a device have been fulfilled". A maker whose test needs no notified body declares conformity itself: it draws up the test's technical documentation and issues an EU declaration of conformity, which states that the regulation's requirements "have been fulfilled". Once the assessment its class requires is complete, the maker affixes the CE marking, by which it "indicates that a device is in conformity with the applicable requirements", before the test is placed on the market. Where a notified body took part, its identification number follows the marking.

Regulation (EU) 2024/1860 amended the IVDR's transitional rules in 2024. It lists the tests whose makers could declare conformity themselves under the old Directive, 98/79/EC, and which now need a notified body: tests in classes D, C and B, and class A tests supplied sterile. Class A tests that are not sterile are the only ones missing from that list, and under Article 48 of the IVDR their makers declare conformity themselves. For a sterile class A test, Article 48 limits the notified body's part to establishing, securing and maintaining the sterile conditions. Under the IVDR's own rules a class C omics test therefore passes through a notified body, and the chapter's final section sets out the transition for tests already on the market under the Directive.

A test's clinical evidence is assembled in its performance evaluation, the subject of the next section. The plan for surveillance after sale that Chapter 20 describes is part of the test's technical documentation. The certificates that notified bodies issue are now kept in EUDAMED, the EU's database for devices. Its module for notified bodies and certificates became mandatory on 28 May 2026, at the same time as three others, for registering manufacturers and other actors, for registering devices by their unique identifiers, and for market surveillance. For records that predate the modules, the HPRA, Ireland's regulator, gives two later dates: 28 November 2026 for registering devices already on the market, and 28 May 2027 for entering certificates issued earlier.

54 — IVDR classes, and who assesses each

Focal detail

Performance evaluation, and the rung it leaves out

The IVDR defines performance evaluation as "an assessment and analysis of data to establish or verify the scientific validity" of a test's analyte, together with the test's analytical performance and, where applicable, its clinical performance. Each of the three elements has its own definition. The scientific validity of an analyte is "the association of an analyte with a clinical condition or a physiological state". Analytical performance is "the ability of a device to correctly detect or measure a particular analyte". Clinical performance is "the ability of a device to yield results that are correlated with a particular clinical condition". The Medical Device Coordination Group, whose guidance documents interpret the regulation, builds the clinical evidence for a test on these three elements in its guidance of January 2022. It calls performance evaluation "a continuous process", so the evaluation is updated through the test's life.

The three elements map onto the ladder of Chapter 22 only in part. Analytical performance matches the first rung, analytical validity, and the CLSI studies of Chapter 19 supply the evidence for it under either name. The second rung, clinical validity, is split in two. Scientific validity asks whether the analyte is associated with the condition at all, and clinical performance asks whether this test's own results correlate with the condition. The third rung, clinical utility, has no element of its own among the three. None of them asks whether acting on a result improves outcomes for patients, although the regulation does ask the maker to confirm, through post-market performance follow-up, that the balance of benefit and risk stays acceptable over the device's lifetime (Chapter 20).

The split matters most for an omics signature. A methylation signature for cancer screening needs evidence that its pattern is associated with cancer, and separate evidence that the test's own results, from its own specimen, assay and pipeline, correlate with cancer in the people it is meant to screen. The signatures of Chapter 22 that failed on independent samples showed an association only in the samples used to find them, so on new samples they had neither kind of evidence. The IVDR's performance evaluation asks for both, and neither is evidence of the third rung.

A test made inside a health institution

The IVDR exempts some tests that a health institution makes and uses itself. Under Article 5(5), such in-house devices are exempt from the regulation's requirements, except the relevant general safety and performance requirements of its Annex I, when the article's conditions are met. The article sets nine conditions, and the first is that the device is not transferred to another legal entity. The second is that it is made and used under an appropriate quality management system, and the third is that the institution's laboratory is compliant with EN ISO 15189, the laboratory standard of Chapter 19, or with national provisions. The fourth condition requires the institution to justify that the needs of the patient group it serves "cannot be met, or cannot be met at the appropriate level of performance by an equivalent device available on the market". Regulation (EU) 2022/112 deferred that condition, and Regulation (EU) 2024/1860 moved its date to 31 December 2030, so as of October 2026 an in-house test must meet the other eight.

The last five conditions concern disclosure and records. The institution gives information to its competent authority on request, makes a public declaration about the devices, keeps documentation for any class D device, makes each device according to that documentation, and reviews the experience gained from the devices' clinical use. The exemption therefore asks for the two systems of Chapter 19, a quality system and a laboratory that complies with ISO 15189 or with national provisions, and leaves out the notified body and the certificate.

Set beside the US route for a laboratory's own test, the exemption is narrower and asks more. In the United States a laboratory establishes the analytical performance of its own test under CLIA, and CMS checks that work at a survey after testing has begun (Chapter 25). An EU in-house test must still meet Annex I's requirements and comes from a laboratory compliant with ISO 15189 or with national provisions. From 31 December 2030 it will also need a justification that no equivalent test on the market meets the patients' needs at the appropriate level of performance. The exemption also applies only inside the institution, because the test cannot pass to another legal entity. A law firm's summary of the Commission's proposal of December 2025 describes three changes to these conditions: they would become more flexible, the condition about an equivalent device would be removed, and central laboratories that make tests only for clinical trials would be added.

Transition dates, and a proposal still under examination

Tests that were on the market under the old Directive, known as legacy devices, move to the IVDR on a timetable that Regulation (EU) 2024/1860 sets. A test that holds a certificate under the Directive may be placed on the market until 31 December 2027. A test whose maker declared it conforming under the Directive, and which now needs a notified body, may be placed on the market until 31 December 2027 in class D, 31 December 2028 in class C, and 31 December 2029 in class B or as a sterile class A test. Each date holds only "provided the conditions set out in paragraph 3c of this Article are met". The conditions begin with a quality management system, which the maker had to have running by 26 May 2025. The maker must also apply formally to a notified body, by 26 May 2025 for class D, 26 May 2026 for class C and 26 May 2027 for class B and sterile class A, and must hold the notified body's written agreement by 26 September of the same year.

The last of the class C deadlines passed on 26 September 2026. A legacy class C test whose maker had no signed agreement with a notified body by that day does not meet the conditions for the extension to 2028. The class D deadlines passed in May and September 2025, and as of October 2026 the class B deadlines lie ahead, in May and September 2027.

55 — The IVDR transition for tests already on the market

ReadoutFocal detail

On 16 December 2025 the European Commission proposed one regulation to amend the IVDR, the regulation on medical devices and two other regulations, "as regards simplifying and reducing the burden of the rules on medical devices and in vitro diagnostic medical devices". The same law firm's summary describes what the proposal would change for tests. Beyond the in-house conditions, performance studies that involve only routine blood draws would no longer need prior authorization, sterile class A tests would no longer need a notified body, and the five-year limit on a certificate's validity would go. Most of the changes would apply six months after the regulation enters into force. The same summary notes that its provisions may change during the legislative process.

The Council and the European Parliament were still examining the proposal separately as of October 2026, and the two institutions had not begun to negotiate a joint text. The Council's working party began to examine it article by article on 23 March 2026. In a progress report dated 5 June 2026, the Council's Presidency reported that "additional work with a view to refining and complementing the proposal will be necessary to reach a General Approach", the position the member states must agree before talks with the Parliament. The working party was due to discuss the Presidency's third compromise text on 26 and 27 October 2026. In the Parliament the public health committee, SANT, leads, and the same report gives early 2027 as the Parliament's aim for a vote on its position. Two law firms report that the Commission expects adoption by the second quarter of 2027. The European Economic and Social Committee gave its opinion on 29 April 2026, by 191 votes to none with 4 abstentions, and the opinion does not bind either institution.

+ The questionUnder India's rules, what turns a reagent into a licensed diagnostic, and which bodies answer for the laboratory that runs a test and for the research behind it?

The class and the place of manufacture decide who licenses

A class B test kit made in India is licensed by the authority of the state in which it is made, while the same kit made abroad is licensed by India's central regulator, and a class C kit is licensed centrally wherever it is made. India regulates in vitro diagnostics, or IVDs, under the Medical Devices Rules, 2017, and the Rules divide the licensing that way. CDSCO, the Central Drugs Standard Control Organisation, classifies these tests from class A to D by risk, and it publishes the classification in lists of IVDs dated 23 July 2021 and, as updated, 25 October 2023. On 6 February 2026 it added a provision for the risk classification of IVDs on its online system for medical devices.

The license to make a test follows its class. A maker of a class A or B test applies to the State Licensing Authority, the licensing body of the state where the test is made, on Form MD-3 and receives its license on Form MD-5. A maker of a class C or D test applies to the Central Licensing Authority, which is CDSCO, on Form MD-7 and receives its license on Form MD-9. Imports of every class go to the Central Licensing Authority, with the application on Form MD-14 and the license issued on Form MD-15. The Rules also provide test licenses, on Forms MD-12 and MD-13 for manufacture and MD-16 and MD-17 for import, for small quantities made or imported for clinical investigation, test, evaluation, demonstration or training.

A new IVD, one that CDSCO has not yet approved for manufacture or import, takes its own route through the Rules. Its sponsor applies on Form MD-24 for permission to run a clinical performance evaluation, a study of the test's performance on specimens from human participants, and CDSCO grants the permission on Form MD-25. The maker or importer also applies on Form MD-28 for permission to make or import the new IVD, which CDSCO grants on Form MD-29 before it issues the license. Other devices take the parallel route of a clinical investigation, on Forms MD-22 and MD-23, and a permission for a device without a predicate, on Forms MD-26 and MD-27.

In the terms of this part, the licensing authority does for an Indian test what the FDA does for a US device and a notified body for an EU test: its decision comes before sale, and in India the class and the place of manufacture decide which authority makes it. A sequencing panel made in India for a class C purpose goes to CDSCO, as does every sequencing panel imported, whatever its class. CDSCO publishes the classes in its lists. Its updated list of 25 October 2023 places a gene analyzer, an instrument that analyzes the sequence of nucleic acids, in class A, and software that interprets human genome data from molecular genetic testing in class C, but it names no sequencing panel. A maker therefore looks for its kind of test in the lists and notes the date of the list before it plans the license, and a maker whose test is not listed applies to CDSCO for its class through the online system.

56 — Who licenses an IVD in India

Focal detail

Licensing dates, and the tests notified earlier

Some IVDs had been notified earlier, which means that a government notification under the Drugs and Cosmetics Act had already brought them into licensing. Among them were blood-grouping reagents and the tests for HIV, hepatitis B surface antigen and hepatitis C, and those were licensed before the dates that follow. For the rest, licensing came in two steps. For devices not already notified, licensing became mandatory for classes A and B on 1 October 2022, following the health ministry's notification of 11 February 2020. It became mandatory for classes C and D on 1 October 2023, under a CDSCO notification of 12 April 2023. CDSCO's answers to frequently asked questions about IVDs give registration for classes C and D until 30 September 2023 and licensing from the next day. Since 1 October 2023, therefore, a test in any class needs a license before it is made or imported for diagnosis.

57 — India: licensing steps and the documents of 2025 and 2026

Focal detail

Research use only, decided by intended use

Products for research sit outside the Rules on one condition. CDSCO's answers to frequently asked questions about IVDs state that products for research use only are not regulated under the Rules provided they are "not intended to be used in human or animals for diagnosis of any disease or disorder". The condition is the product's intended use. The same reagent, instrument or software is outside the Rules while it is intended for research, and inside them, in the class its diagnostic purpose sets, once it is intended for diagnosis. On 13 March 2026 CDSCO added a second addendum to the current edition of these answers. The addendum does not mention research use, and it answers procedural questions instead, among them that the applications on Forms MD-24 and MD-28 may be filed together and that each change to the version of approved software needs an application for a post-approval change (Chapter 20).

The United States draws the line in the same place and marks it on the label. The FDA's labeling regulation requires a product in the research phase to state on its label that it is "Not for use in diagnostic procedures" (Chapter 25). For omics the line matters because the same sequencers, kits and analysis programs serve research and diagnosis. A team that moves a research method toward patients therefore moves it across this line, and in India a reagent, instrument or program made or imported for diagnosis then needs a license in its class.

Software, performance evaluation and the drafts of 2026

CDSCO issued its guidance document on medical device software under the Rules by circular on 21 July 2026. The document itself carries its number but no date. CDSCO's list of public notices dates it 21 July 2026, as do a law firm and two consultancies that have reported on the guidance under the same document number. The guidance covers software that meets the definition of a medical device, and CDSCO states that its term for such software includes software for in vitro diagnostics unless otherwise specified. Software that drives or influences a hardware device takes the class of that device. Standalone software is classed on two scales: the state of the healthcare situation, critical, serious or non-serious, and the significance of the software's information to the healthcare decision, from treatment or diagnosis down to informing clinical management. The combinations run from class D down to class A.

For changes to software, the guidance asks that change management be documented as part of the risk management plan. It states that an Algorithm Change Protocol "may be devised, wherever applicable, based on the nature and risks", so the protocol is an option, not a requirement. The protocol it describes has five plans, for data management, for performance evaluation and monitoring, for retraining the algorithm where that applies, for software updates and for rollback. CDSCO states that the guidance "should not be misconstrued as a new regulatory control" on such software. One of the consultancies dates the draft that preceded it to 2025. The guidance reaches the software of Chapter 19, the aligners, variant callers and pipelines that do much of an omics test's measuring, when that software meets the definition of a device.

CDSCO's page on IVDs also lists a series of documents on performance evaluation. On 17 September 2025 it published standard protocols for evaluating the performance of IVDs, prepared with ICMR, the Indian Council of Medical Research. The next day it launched a compendium of those protocols and a portal for validating IVD kits, a launch that CDSCO describes as a "Global first initiative". On 29 June 2026 it updated its list of laboratories for performance evaluation. Two guidances of 2026 are drafts for comment as of October 2026: one for the import of IVDs, dated 30 January 2026 and posted on 2 February 2026, and a "Draft Guidance document for Manufacturers of In-vitro diagnostic Medical device for stakeholders comment", posted on 9 October 2026.

An assessment outside the three jurisdictions, WHO's prequalification of in vitro diagnostics, covers listed analytes only. Its eligibility tables, last updated on 27 September 2026, name tests for HIV, hepatitis B and C, malaria, syphilis, tuberculosis, HPV by nucleic-acid testing, G6PD, SARS-CoV-2, point-of-care glucose, HbA1c and hemoglobin, and gonorrhea, chlamydia and trichomoniasis. The only sequencing entry is targeted sequencing for drug resistance in tuberculosis. No genomic, multi-omics or protein signature test is eligible, and no cancer screening other than for HPV. The tables list what may be assessed, not tests that WHO has prequalified.

Laboratories, accreditation and research ethics

Chapter 19 described NABL's move of India's medical laboratories to ISO 15189:2022. NABL announced the move on 29 December 2022 and set 31 December 2025 as the deadline for its criteria NABL 112A. Between those two dates, NABL issued an addendum to its earlier criteria, NABL 112, for the new edition on 13 February 2024, and on 18 December 2024 it published NABL 112A with its guidance document, NABL 112B. ILAC, the international cooperation of laboratory accreditation bodies, had set a transition of three years from the standard's publication on 8 December 2022, which also ended in December 2025. A laboratory that cites the 2012 edition in October 2026 therefore cites an edition that ISO has superseded and NABL no longer accredits against.

ISO 15189 appears in two of this part's jurisdictions in different roles. In India it is the standard against which NABL accredits medical laboratories, and in the EU a health institution's laboratory must comply with it, or with national provisions, before the institution may use its own tests under the in-house exemption (Chapter 26). Some Indian genomics laboratories also state that they hold accreditation from CAP, the US body of Chapter 19 (Chapter 23).

Research on people follows a separate set of rules. ICMR issued its National Ethical Guidelines for Biomedical and Health Research Involving Human Participants in October 2017. Their scope is research for health conducted in India that involves human participants. Their Section 10, on human genetics testing and research, has subsections on genetic counseling, informed consent, the results of genetic testing, and genetic diagnosis, testing and screening. An omics study that recruits participants in India therefore works under them, from consent to the handling of results.

Prenatal diagnosis is governed by a statute of its own, the Pre-Conception and Pre-Natal Diagnostic Techniques (Prohibition of Sex Selection) Act, 1994. The Act defines a prenatal diagnostic test to include any test or analysis of the blood of a pregnant woman conducted to detect genetic or metabolic disorders or chromosomal abnormalities. The words of that definition cover a cell-free DNA test of maternal blood, such as the prenatal screen of Chapter 14, and a search for this guide found no court ruling or ministry statement on such tests. A genetic laboratory, genetic clinic or genetic counseling center may take part in prenatal testing only once it is registered under the Act. The Act allows the tests only to detect the abnormalities it lists, and only when the person conducting them records in writing that one of its conditions is met, such as the pregnant woman being above 35 years of age. It prohibits telling the woman, her relatives or anyone else the sex of the fetus.

Seven percent, and the denominator a regulated figure carries

The first reports of the opening's 7 percent counted it against the wrong whole: NASA's page of January 2018 put it in terms of the astronaut's genes, and the wire story of March in terms of his DNA. NASA's statement of 15 March 2018 named the right whole, the gene expression that had changed during spaceflight (Chapter 7). A share whose whole is wrong or missing can be read against any whole the reader supplies, and in 2018 it was read against three.

The records of this part show what a regulated test's label and the documents that govern it state in place of a bare share. A US product still under test carries on its label the sentence "The performance characteristics of this product have not been established" (Chapter 25). The FDA's summary of the 2024 colorectal approval gives its sensitivity and its false-positive rate each with the group in which it was counted, and with the number of people studied (Chapter 22). The FDA's guidance for sequencing tests asks the developer to set each accuracy threshold on the lower bound of a 95 % confidence interval as well as on the point estimate, and that bound moves with the number of samples behind it (Chapter 19). The IVDR defines analytical and clinical performance as abilities "of a device", so its figures belong to one test, with its own specimen and pipeline (Chapter 26). The investigational label states plainly that no figure has yet been established, and the FDA's summary, its sequencing guidance and the IVDR's definitions fix the whole against which a figure is counted.

The 7 percent also holds the question this guide began with. What the flight changed was a state of the cells, the expression of genes. The genome does not record that state, and the inherited sequence did not change (Chapters 1 and 15). That state could be counted because RNA can be copied and paired, the property that makes the nucleic-acid layers the easiest to measure and leaves proteins and metabolites the hardest (Chapter 2). Turning such a count into a result someone can act on takes the rest of the book. It takes a chain from tube to report whose every link is controlled, the subject of Chapter 17, and a claim stated exactly, proven against traced evidence and kept proven after release, the subjects of Chapters 19 and 20. It takes a climb from analytical through clinical validity to clinical utility, which Chapter 22 describes. It also takes the review that each jurisdiction of this part sets for the route a test takes, by the FDA or CMS, a notified body or a licensing authority. Each of those steps fixes something the 7 percent left open when it was first read: what was measured, in whom, by which version of the test, and against what whole.

The chapters reduce to a short set of working rules for anyone who designs, runs, buys or reviews omics measurements. Each one traces back to a mechanism explained earlier.

1. Name the layer before the technology. Decide whether the question concerns the code, its accessibility, its use, its products or its chemistry, and let that choose the method (Chapters 1 and 15).
2. Ask where the signal per molecule comes from and how identity is assigned. Amplification, a bright label or single-molecule detection supplies the signal, and sequence, mass or a tested binder assigns identity, so the answers predict breadth, depth and failure modes (Chapters 2 and 3).
3. Specify minimums where they matter, not averages. State the coverage across the target regions, the read depth at each CpG and the events in each gate as minimums, because a call, a methylation level or a population count is only as precise as the reads or events behind it (Chapters 4, 6 and 11).
4. Keep aliquots for orthogonal confirmation by a targeted method before any finding drives a decision (Chapters 7, 9 and 10).
5. Treat every name in a results table as a claim with a confidence level. Protein groups are inferred, metabolite annotations carry identification levels, and clusters are interpretations that need independent evidence (Chapters 8, 10 and 12).
6. State the detection limit with every count or concentration, in calibrated units, with the gates and controls that produced it (Chapters 11 and 14).
7. Push something known through the chain. Spike-ins, reference materials and calibration beads test every conversion from the point where they enter, and only where their values are known (Chapters 16 and 17).
8. Map the test from the tube to the signed report. Give each link an owner, a format at its interface and a control, and write the tube type, the time to centrifugation and the storage conditions into the specification, because the specimen changes before any instrument measures it (Chapter 17).
9. Balance groups across batches and include pooled quality controls. No analysis can separate a comparison from the processing batches it was confounded with (Chapter 18).
10. Read fold change and adjusted significance together, never either alone (Chapter 18).
11. Version the software as part of the measurement. Record the version with every result, lock the pipeline before validation, and treat an update to any of its tools, open-source ones included, as a change to the test (Chapters 16, 19 and 22).
12. Prove the test against a claim stated exactly. Trace every requirement to the study that proves it, and accept the risk left after the controls only against criteria written into the risk management plan before the analysis and, taken as a whole, against the benefit (Chapter 19).
13. Keep the proof current after release. Control suppliers in proportion to risk, review every change against the trace before making it, and treat each complaint as a sign that the evidence may no longer hold, because the clocks for reporting a wrong result start when the maker learns of it (Chapter 20).
14. Validate a locked signature on samples it has never seen before anyone acts on it (Chapter 22).
15. Write the intended use first: population, specimen, analyte, result and decision. The evidence plan and the regulatory route follow from it (Chapters 22 and 25 to 27).
16. Read a maker's figure as the maker's claim. Specifications, prices, counts and claims to be first stay the maker's own unless a regulator's record supports them, and because owners, names and records change within months, every check carries its date (Chapter 23).
17. Keep demonstrated and announced apart in every claim, including the team's own (Chapter 24).
18. Decide kit or laboratory test market by market. In the United States a test sold as a device takes the FDA route its class sets, while a laboratory's own test falls under CLIA, whose surveys check its analytical performance after testing has started. In the EU an in-house test is exempt only under the conditions of Article 5(5), and in India a kit made or imported for diagnosis needs a license from the authority that its class and place of manufacture decide (Chapters 25 to 27).

Terms are defined as they are used in this guide.

16S ribosomal RNA gene

A gene that every bacterium carries, about 1,500 bases long, with parts nearly identical across bacteria and nine hypervariable regions that differ between lineages; amplified and sequenced, it shows which bacteria a sample holds.

180-day supplement

The PMA supplement the FDA requires for a significant change in design, specification or software of an approved device, which it reviews in depth and approves before the change is made.

30-day notice

The route for a change to the manufacturing methods of an approved device: the maker may make the change 30 days after the FDA receives the notice, unless the FDA finds the notice inadequate.

510(k) (premarket notification)

A submission showing that a device is substantially equivalent to a legally marketed predicate not subject to premarket approval; the FDA's decision on it is a clearance.

Abundance

The number of copies of a molecule in a sample.

Adapter

A short synthetic DNA sequence attached to the ends of each fragment so that the sequencer can grip and read it.

Advisory panel (FDA)

A committee of outside experts that discusses an application in public and votes; its vote advises the FDA without binding it.

Affinity reagent

A molecule, usually an antibody or aptamer, that binds a target by shape. Its specificity must be shown for each target.

Algorithm Change Protocol

In CDSCO's software guidance of July 2026, an optional protocol for changes to medical device software, which a maker may devise where applicable as part of the software change management documented in its risk management plan. It has five plans: data management, performance evaluation and monitoring, retraining where it applies, software updates and rollback.

Alignment

Placing each sequencing read at the position in a reference genome that it best matches.

Analyte

The substance a test is meant to measure.

Analytical performance (IVDR)

The ability of a device to correctly detect or measure a particular analyte.

Analytical sensitivity

An instrument's or assay's ability to detect a weak signal; a different quantity from diagnostic sensitivity, the share of true cases that a test detects.

Analytical validation

The studies that show a test measures its analyte correctly: as applicable, accuracy, precision, analytical sensitivity, analytical specificity including interfering substances, the reportable range and the reference intervals.

Analytical validity

Whether a test measures its analyte accurately and reproducibly in the specimens it will receive.

Aptamer

A short single strand of DNA or RNA, often chemically modified, selected to fold into a shape that binds a particular target.

ATAC-seq

A method that maps open chromatin by letting an adapter-inserting enzyme act only where DNA is accessible.

Authentic standard

A pure sample of a suspected compound, run under the same conditions as the sample to show that the two match on at least two independent properties.

Base calling

The conversion of a sequencer's raw signal, light or current, into letters with quality scores.

Batch

A set of samples processed together, which leaves a small systematic shift on every measurement.

Batch effect

The systematic shift that laboratory conditions, reagent lots and personnel leave across samples processed together; across thousands of features it can be larger than the biology, and a balanced design is its control.

Becoming aware (US reporting)

The start of an FDA reporting clock. For a 30-day report, and for a 5-day report the FDA has requested, a maker becomes aware when any employee learns of information that reasonably suggests a report is required. For a 5-day report on an event that needs remedial action, it becomes aware when an employee in a particular role learns that the action is needed: FDA's guidance names a manager or supervisor of staff with regulatory, scientific or technical duties, and its wording leaves open whether an employee whose duties concern the collection and reporting of adverse events also counts.

Biomarker

A molecule whose level indicates a state of the body, such as a disease or a response to treatment.

BIOSECURE Act

A US law enacted on 18 December 2025 within the defense authorization act for 2026, whose prohibitions on biotechnology companies of concern take effect only after the federal acquisition rules are revised.

Bisulfite conversion

A chemical treatment that turns unmethylated cytosine into uracil, read as T, while methylated cytosine stays C.

Bottom-up proteomics

Proteomics in which proteins are first digested, usually with trypsin, into peptides that are then measured.

Breakthrough Device designation

A designation the FDA grants, under a voluntary program, to some devices that provide more effective diagnosis or treatment of a life-threatening or irreversibly debilitating disease and meet at least one further criterion, such as having no approved or cleared alternative. It brings interaction with FDA experts and prioritized review, and the device must still meet the FDA's standards; a designation is not a clearance, a grant of a De Novo request or an approval.

cDNA

Complementary DNA: a DNA copy of RNA made by reverse transcriptase, so that RNA can be amplified and sequenced.

CDSCO

The Central Drugs Standard Control Organisation, India's national regulator for drugs and devices; under the Medical Devices Rules it is the Central Licensing Authority and publishes the IVD classification lists.

CE marking

The marking by which a maker indicates that a device conforms to the applicable requirements of the IVDR. It is affixed before the device is placed on the market, once the conformity assessment its class requires is complete, and the notified body's identification number follows it where one took part.

Cell barcode

A DNA sequence shared by every molecule captured from one cell, used to sort pooled reads back into cells.

Cell-free DNA (cfDNA)

Short fragments of DNA in plasma, released when cells die and mostly about 166 base pairs long, about the length of DNA on one nucleosome.

Central Licensing Authority

Under India's Medical Devices Rules, CDSCO in its role of licensing the manufacture of class C and D devices and the import of devices of every class.

CFIUS

The Committee on Foreign Investment in the United States, which reviews foreign investment in US businesses.

Change control

The process through which every change to a released design passes before it is implemented: the change is reviewed for its effects, and the studies it touches are run again as the review requires. ISO 13485 requires design and development changes to be controlled before they are implemented (clause 7.3.9).

Change review (design change)

The step of change control that judges what a change affects before it is implemented; under clause 7.3.9 of ISO 13485 it includes evaluating the change's effect on the device's constituent parts, and the trace of the design shows which studies the change requires to be run again.

Chromatin accessibility

How far a stretch of DNA is free of nucleosomes and other proteins, and so reachable by the transcription machinery.

Circulating tumor DNA (ctDNA)

The fraction of cell-free DNA released by a tumor.

Clearance

The FDA's decision on a 510(k). A cleared device is not described as approved.

CLIA

The US law for clinical laboratories, applied through a program that CMS runs. It certifies laboratories, sorts tests by complexity and requires a laboratory to establish the analytical performance of a test it builds or modifies; it does not address clinical validity.

CLIA complexity

The category of a test under CLIA: waived, moderate or high. FDA scores a test system on seven criteria from 1 to 3; 12 or less is moderate, and a test system not yet categorized defaults to high.

Clinical performance (IVDR)

The ability of a device to yield results that are correlated with a particular clinical condition.

Clinical performance evaluation

Under India's Medical Devices Rules, a study of a new IVD's performance on specimens from human participants, run with CDSCO's permission (applied for on Form MD-24, granted on Form MD-25).

Clinical utility

Whether acting on a test result improves outcomes compared with not testing.

Clinical validity

Whether a test result corresponds to the condition of interest in the intended population.

Cluster (cell data)

A group of cells with similar measurements, formed by an algorithm; a statistical grouping, so calling it a cell type needs independent evidence.

Co-isolates

Materials isolated along with extracellular vesicles, such as lipoproteins, platelets or soluble proteins; reporting how far EVs were enriched over them is how a study states its preparation's purity.

Coefficient of variation (CV)

The spread of repeated measurements as a percentage of their mean, measured on a pooled quality-control sample run at intervals through a batch.

Companion diagnostic

A test essential for the safe and effective use of a particular drug. The IVDR defines it as a device essential for the safe and effective use of a corresponding medicinal product.

Compensation

The mathematical correction of fluorescence spillover between detectors in flow cytometry.

Competent authority (EU)

A member state's regulator for devices; under the IVDR it receives the maker's reports of serious incidents and evaluates the risks.

Complaint record (QMSR)

The maker's record of a complaint under 21 CFR 820.35: the device's name, the date received, the unique device identifier or other identifiers, the complainant's contact details, the nature and details of the complaint, any corrective action and reply, and the records of review and investigation, with a justification when no investigation was done.

Conformity assessment

In the IVDR's words, "the process demonstrating whether the requirements of this Regulation relating to a device have been fulfilled"; a notified body takes part for every class but non-sterile class A, whose makers declare conformity themselves.

Confounded

Of two factors that cannot be separated in the data, such as treatment and processing day when every treated sample was processed on one day and every control on another.

Control

A known quantity passed through every step alongside the samples, so that anything lost or distorted shows up as a difference between what went in and what came out.

Copying

The first of the two properties that make nucleic acids easy to measure: enzymes copy DNA, and RNA once it has been written into DNA, so a few molecules can be multiplied until an instrument detects them.

Correction (21 CFR 806)

The repair, modification, adjustment, relabeling, destruction or inspection of a device, including patient monitoring, without moving it from its point of use. A maker or importer reports one to the FDA when it is made to reduce a risk to health or to remedy a violation of the law that may present one.

Corrective action

In ISO 13485 (clause 8.5.2), action on a nonconformity so that it does not recur, proportionate to the effects of the nonconformity. The IVDR defines it more widely, as action to eliminate the cause of a potential or actual non-conformity or other undesirable situation.

Count table

The output of RNA-seq: genes down one side, samples across the top, and in each cell the number of sequenced fragments, not molecules.

Coverage

The number of sequencing reads overlapping a position, or its average across a region or genome, written as 30× and so on.

CpG

A cytosine followed by a guanine in the DNA sequence: the usual site of DNA methylation in human cells.

Critical quality attribute

A property of a product that should stay within set limits to ensure its quality, such as the glycan pattern of a therapeutic antibody.

Cross-reactivity

Accidental binding, such as a detection antibody meeting the wrong capture antibody in a mixture, which produces false signal and limits how many antibody pairs a conventional multiplexed immunoassay can mix.

Cytomics

The measurement of many properties of individual cells across very many cells.

Data-dependent acquisition (DDA)

A mass spectrometry mode that selects the most intense ions in each MS1 scan for fragmentation; the choice is partly random, so a peptide measured in one run may be skipped in the next.

Data-independent acquisition (DIA)

A mass spectrometry mode that fragments all ions within successive mass windows rather than selecting the most intense.

De Novo request

A route for classifying a novel device that has no predicate, where general controls, alone or with special controls, give reasonable assurance of safety and effectiveness; the FDA grants the request.

Demultiplexing

Sorting the reads of a pooled sequencing run back to their samples by their indexes.

Design and development file

The record kept for each device type or family that includes or references the evidence of conformity to the design requirements.

Design controls

The design and development requirements of ISO 13485 clause 7.3: planning, inputs, outputs, review, verification, validation, transfer, change control and the design and development file.

Design input

A requirement the design must meet, complete, unambiguous, able to be verified or validated, and taking account of risk management outputs.

Design output

A protocol, software, specification or label that embodies the design; it must meet the inputs and contain or reference acceptance criteria.

Detection limit

The smallest amount or signal an instrument or assay can distinguish reliably from background.

Device class (US)

One of three classes in 21 CFR 860.3: class I under general controls only, class II under special controls, and class III, for which premarket approval is required.

Differential expression

The comparison of each gene's level between conditions; the normalization methods used for it estimate a scaling factor from the genes that did not change.

Differential result

A table that compares two groups of samples feature by feature, giving for each feature a fold change, a p-value, the p-value adjusted for the number of features tested, and the feature's abundance.

DNA methylation

A small methyl group attached to cytosine, in most cell types almost always at a CpG site, and copied through cell division.

Doublet

A droplet that captured two cells, which looks in the data like a hybrid cell type that does not exist.

Duplex mode (sequencing by expansion)

The sequencing-by-expansion mode in which the two strands of each molecule are linked during library preparation, so that their reads can be combined for accuracy.

Dynamic range

The ratio between the largest and smallest quantities present in a sample, or measurable by an instrument in one run.

Electrospray

The usual way of ionizing proteins and peptides for mass spectrometry: liquid from a chromatography column is sprayed from a fine needle at high voltage, the charged droplets shrink as the solvent evaporates, and the dissolved molecules are left carrying extra protons.

Epigenome

The chemical marks on DNA and on the proteins that package it, which record which genes a cell keeps open for use, often through many divisions.

EUDAMED

The EU's database for medical devices and IVDs; four of its modules, among them the one for notified bodies and certificates, became mandatory on 28 May 2026.

Event

In cytometry, a pulse that rises above the instrument's threshold, usually one cell, recorded as one row of measurements.

Exome

The protein-coding regions of the genome, less than 1.5 % of it, which can be sequenced alone.

Exon

A segment of a gene that is kept in a messenger RNA; a cell can join different subsets of a gene's exons into different messenger RNAs.

Expected erroneous result (IVDR)

A wrong result of the kind that the product information clearly documents and quantifies. It is left out of serious-incident reports and counted instead in trend reporting.

Expression (of a gene)

A gene is expressed when it is transcribed, and its expression is usually measured as the amount of its RNA, or of the protein made from it.

Expression signature

A panel, a fixed set of genes, measured in a tumor to estimate the risk that the cancer will return.

Extracellular vesicle (EV)

A particle enclosed by a lipid bilayer, released by cells, carrying protein, RNA and lipid cargo.

Failure modes and effects analysis (FMEA)

A risk analysis that starts from the parts and steps of a design or process, asks how each might fail and follows each failure to its effects; IEC 60812 describes it.

False discovery rate (FDR)

The expected share of false findings among all results declared significant.

FASTQ

The common file format for sequencing reads, holding each read's sequence with a quality score for every base.

Fault tree analysis

A risk analysis that starts from one undesired event, such as a false-positive report, and works back through the events and failure modes that could combine to produce it; IEC 61025 describes it.

Feature

Anything a study measures, such as a gene, a protein or a metabolite. In untargeted metabolomics, a signal defined by a mass-to-charge ratio and a retention time, which may or may not be a distinct compound.

Field safety corrective action (FSCA)

Under the IVDR, a corrective action that a maker takes for technical or medical reasons to prevent or reduce the risk of a serious incident; it is reported before it is undertaken, except in urgent cases.

Field safety notice

Under the IVDR, the maker's communication to users or customers about a field safety corrective action; it must explain the risk clearly without understating it, state every action users must take, and be publicly accessible.

Fold change

The ratio of a quantity between two conditions, usually reported on a log₂ scale.

Forward scatter

Light scattered by a cell at small forward angles, which grows with the cell's size.

Foundation model (single-cell)

A model trained on many single-cell datasets at once.

Gate

A boundary drawn on a two-dimensional plot of one measured parameter against another to select a population of cells or particles.

Gating

Selecting a population of cells or particles by drawing boundaries on plots of their measured parameters.

GC-MS

Gas chromatography with mass spectrometry, for molecules that are volatile or can be made volatile by chemical derivatization; its reproducible fragmentation supports large libraries of reference spectra.

General controls

The basic provisions of the US Medical Device Amendments of 28 May 1976, among them those on adulteration and misbranding, registration and listing, premarket notification, records and reports, and good manufacturing practice; they apply to devices in all three classes unless a device is exempted, and class I devices are subject to them alone.

Genome

The complete DNA sequence a person inherits, the same in nearly every cell of the body.

Glycomics

The measurement of glycans, the branched sugar chains attached to proteins and lipids.

Half-life

The time in which half of the molecules present are destroyed.

Harm

In ISO 14971, injury or damage to the health of people, or damage to property or the environment.

Hazard

In ISO 14971, a potential source of harm.

Hazardous situation

In ISO 14971, as Health Canada and FDA reproduce it, a circumstance in which people, property or the environment are exposed to one or more hazards; this guide reads it, for a test, as a wrong result once it reaches a decision about a patient.

Health technology assessment

An evaluation of a medicine, device or test that informs a health system's decisions on paying for it. In the EU, Regulation (EU) 2021/2282 has applied since January 2025: it provides for joint clinical assessments, for which tests of the highest risk class may be selected, and leaves pricing and reimbursement to each member state.

Hemolysis

The bursting of red cells in a blood specimen, which releases their contents into the plasma.

Heterozygous

Of a variant: present on only one of the two parental copies of a chromosome, so that it appears in roughly half the reads at its position.

Histone modification

A chemical mark on the histones, the proteins that make up each nucleosome, that recruits or repels the transcription machinery.

Homozygous

Of a variant: present on both parental copies of a chromosome, so that it appears in nearly all the reads at its position.

Hydrodynamic focusing

Narrowing a sample stream inside a faster sheath flow so that cells pass a detector in single file.

Hypervariable regions

The nine parts of the 16S ribosomal RNA gene that differ between bacterial lineages.

In vitro diagnostic product (US)

In FDA's regulation, the reagents, instruments and systems intended for use in the diagnosis of disease or other conditions; the definition turns on intended use.

In-house device (IVDR)

A test that a health institution makes and uses itself without transferring it to another legal entity. Under Article 5(5) of the IVDR it is exempt from the regulation's requirements, except the relevant general safety and performance requirements of Annex I, when the article's conditions are met: eight of the nine apply as of October 2026, among them an appropriate quality management system and a laboratory compliant with EN ISO 15189 or with national provisions, and the ninth, a justification that no equivalent device on the market meets the patient group's needs at the appropriate level of performance, applies from 31 December 2030.

Incident (IVDR)

Any malfunction or deterioration in the characteristics or performance of a device on the market, including use error due to ergonomic features and inadequate information from the maker, and any harm that follows from a medical decision, or an action taken or not taken, on the basis of the device's result.

Index hopping

The appearance of a read carrying the index of a library it did not come from, so that it is assigned to the wrong sample.

Information value

A value NIST characterizes for a reference material without certifying it; it cannot be used to establish metrological traceability.

Instruction set

The genome seen as what constrains what a cell can become, as against its state.

Interface

The place where one link of a test hands its output to the next, often as a file in an agreed format.

Internal standard

A known amount of an isotope-labeled copy of the measured molecule added to a sample, which turns signal into an absolute amount.

Investigational Use Only (IUO)

The US labeling statement for a product tested before full commercial marketing: "For Investigational Use Only. The performance characteristics of this product have not been established."

ISO 15189

The international standard for quality and competence in medical laboratories, in force in its fourth edition of 2022; a basis for NABL's accreditation in India and a condition of the in-house exemption in the EU.

Isobaric tags

Isotope labels that weigh the same, so labeled samples are measured together and told apart by reporter ions released on fragmentation.

Isomers

Molecules with the same atoms, and therefore the same mass, arranged differently.

IVDR

Regulation (EU) 2017/746, the European Union's In Vitro Diagnostic Regulation, which sorts tests into classes A to D by the rules of its Annex VIII.

Laboratory-developed test (LDT)

A test that a single laboratory designs and runs itself rather than one sold as a kit; in the United States it falls under CLIA.

Layer

One whole set of molecules of one kind that omics measures, such as the genome, the transcriptome or the proteome.

LC-MS

Liquid chromatography coupled to mass spectrometry, which covers the widest range of metabolites.

Legacy device (IVDR transition)

A test placed on the market under the old Directive 98/79/EC, which may stay on the market until a date set by class in Regulation (EU) 2024/1860, provided its maker met the regulation's conditions.

Library

The collection of adapter-tagged fragments prepared from a sample for sequencing.

Library preparation

The laboratory steps that turn extracted DNA, or cDNA, into a library: fragmenting it and fitting each fragment with adapters at both ends.

Lipidomics

The measurement of the thousands of lipid species in membranes, stores and signaling pathways, by mass spectrometry.

Liquid biopsy

Testing the material that tissues release into the blood, such as cell-free DNA and extracellular vesicles, as a sample of tissues otherwise reachable only by a tissue biopsy.

Locked

Fixed before validation, so that what runs on patients' samples is what was tested: for a pipeline, every tool, version, parameter and reference; for a signature, every step of its processing and its decision threshold.

Log₂ fold change

A fold change counted in doublings: 1 is a doubling, −1 a halving, and every further unit another factor of two.

Long read

A sequencing read from a single native molecule, typically thousands to hundreds of thousands of bases long.

Low-biomass sample

A sample with little microbial DNA of its own, such as blood, spinal fluid or a tissue biopsy, in which traces of DNA from reagents can dominate the result.

Major change (India)

A change listed as major in the Sixth Schedule of India's Medical Devices Rules; the license holder needs the licensing authority's prior approval, and the change is deemed approved if the authority does not answer within 45 days.

Malfunction (US reporting)

The failure of a device to meet its performance specifications or otherwise perform as intended, where the performance specifications include all claims made in the device's labeling. It is reportable when it would be likely to cause or contribute to a death or serious injury if it recurred, and FDA treats the failure of a diagnostic device in the same way.

Marker

A molecule, such as a protein, whose level distinguishes a cell type or state.

Mass cytometry

Cytometry that labels antibodies with metal isotopes and reads them by mass instead of fluorescence.

Mass-to-charge ratio (m/z)

The quantity a mass spectrometer actually measures: an ion's mass divided by its charge.

Materiovigilance Programme of India (MvPI)

India's national program that collects reports of adverse events from medical devices. By a compliance-software firm's account of June 2024, a CDSCO circular directs reports of suspected unexpected serious adverse events to it.

Medical Device Reporting (21 CFR 803)

The US regulation under which makers and importers report deaths, serious injuries and malfunctions, and user facilities report deaths and serious injuries, within set periods after becoming aware of them: 30 calendar days for a maker, or 5 work days when the FDA designates the event or it requires remedial action.

Medical Devices Rules, 2017

The rules under which India regulates medical devices, in vitro diagnostics included, with four risk classes, A to D, and licenses for manufacture and import.

Metabolites

The small molecules of life, typically under about 1,500 daltons: sugars, amino acids, fatty acids and other lipids, nucleotides, vitamins, hormones and the intermediates of every pathway that makes or breaks them.

Metabolome

The full set of small molecules in a cell or sample, such as sugars, amino acids and fats, recording the chemistry that proteins carry out.

Metagenomics

Sequencing the mixed DNA of a microbial community without culturing its members.

Metatranscriptomics

Sequencing the RNA of a microbial community, which shows which genes the community is actively using.

Methylation level

The fraction of reads at a site that carry a methyl mark: a population average, not a property of one cell.

MISEV2023

The International Society for Extracellular Vesicles' guidelines on nomenclature, isolation, characterization and reporting.

Mock community

A sample of microbes of known composition processed alongside real samples as a positive control.

Molecular residual disease

The traces of cancer that remain after treatment and can be detected only by molecular tests.

MS1 and MS2

The two spectra of tandem mass spectrometry: MS1 records the ions present at one moment, and MS2 the masses of the fragments of selected ions broken by collision with gas.

Multiple testing

The inflation of chance findings when many hypotheses are tested at once.

Multiplexing

Measuring many analytes, such as many proteins, in one well or tube.

mzML

The Proteomics Standards Initiative's open file format for raw mass spectrometer output.

NABL

The National Accreditation Board for Testing and Calibration Laboratories, which accredits Indian medical laboratories against ISO 15189:2022 through its criteria NABL 112A.

Nanoparticle tracking analysis (NTA)

A method that sizes and counts particles by following the scattered or fluorescent light of each as it drifts.

Negative percent agreement (NPA)

The share of a reference's known non-variant positions that a test leaves uncalled.

Nominally significant

Passing the conventional threshold of 0.05 before any allowance is made for the number of endpoints tested.

Non-notified device (India)

A device that had not been notified under the Drugs and Cosmetics Act before the Medical Devices Rules; such IVDs came under licensing on 1 October 2022 for classes A and B and on 1 October 2023 for classes C and D.

Nonconformity

A failure to meet a requirement, such as a result outside a test's specification.

Notified body

A conformity assessment body designated under the IVDR. It assesses tests in classes B, C and D before they are placed on the market, and for a sterile class A test only the sterile conditions.

Nuclear magnetic resonance (NMR)

A method that reads the spin of atomic nuclei: in a strong magnetic field certain nuclei absorb radio waves at frequencies set by their chemical surroundings, so a spectrum reports the atomic environments in a molecule. It is the main exception to the five properties most instruments read.

Nucleosome

A spool of histone proteins around which DNA is wound; a short stretch of linker DNA lies between one spool and the next.

Off-the-shelf software

FDA's term for a generally available software component whose complete life cycle the device maker cannot control.

Omics

The measurement of a complete set of molecules of one kind, a whole layer, in a single experiment.

Outsourced process

Part of a maker's work that an outside party carries out for it. ISO 13485 requires its controls to be proportionate to the risk involved and to the outside party's ability to meet requirements.

Overall residual risk

All the residual risks of a device taken together, weighed against its benefit.

p-value

How often chance alone would produce a difference at least as large as the one observed, if there were truly no difference.

Pairing

The second of the two properties: a strand of DNA or RNA binds a partner strand letter by letter, so a probe for any sequence can be designed from the sequence alone.

Panel

A set of probes, assays, antibodies or genes chosen to be measured together; the expression signatures of Chapter 21 measure a fixed set of genes.

Patterned flow cell

A sequencing flow cell whose surface Illumina, its maker, describes as nanowells at fixed, evenly spaced locations, in which clusters grow by a method it calls exclusion amplification.

PCR-free library

A sequencing library prepared without amplification.

Peptide

A short chain of amino acids; in bottom-up proteomics, a piece of a protein cut by trypsin, typically roughly 7 to 30 amino acids long.

Performance evaluation

Under the IVDR, the assessment and analysis of data that establishes or verifies a test's scientific validity, analytical performance and, where applicable, clinical performance.

Periodic safety update report (PSUR)

Under the IVDR, the post-market surveillance report that a maker prepares for a class C or D device.

Pharmacogenomics

Using a person's genotype to choose a drug or its dose.

Phasing (sequencing)

Loss of synchrony in a sequencing cluster when some strands miss an addition and fall behind the rest; with prephasing, it degrades base quality along a read.

Phasing (variants)

Showing which variants sit together on the same parental copy of a chromosome, which a long read can do.

Phenotype

The observable outcome of a cell or an organism.

Phosphoproteomics

The enrichment and measurement of phosphorylated peptides, to map which proteins in signaling pathways carry a phosphate.

Positive percent agreement (PPA)

The share of a reference's known variants that a test calls.

Positive predictive value

The share of positive test results that are true positives. It depends strongly on how common the condition is.

Post-market performance follow-up (PMPF)

Under the IVDR, the follow-up that confirms, over a device's expected lifetime, that its safety, performance and clinical evidence remain valid and its balance of benefit and risk acceptable; MDCG guidance calls it a continuous process that updates the performance evaluation.

Post-market surveillance

A maker's systematic collection and analysis of information from its devices on the market. Under the IVDR it is a planned system whose plan is part of the technical documentation, with a surveillance report for classes A and B and a periodic safety update report for classes C and D.

Pre-analytical phase

Everything that happens to a specimen before the analytical test: the tube, transport, time to centrifugation, processing and storage. ISO's standards call it pre-examination.

Predetermined change control plan

A plan, allowed under US law since 2022, that sets out in advance what may change in a device after release, such as a model in its software, and how each change will be verified.

Predicate

A legally marketed device, not subject to premarket approval, to which a 510(k) shows substantial equivalence.

Premarket approval (PMA)

The FDA's review of the safety and effectiveness of a class III device; the FDA approves the application, and changes to an approved device can come back to it as supplements.

Prephasing

Loss of synchrony in a sequencing cluster when some strands add two bases in one cycle and run ahead of the rest.

Preventive action

In ISO 13485 (clause 8.5.3), action on a potential problem before it occurs, proportionate to the effects of the potential problem.

Primer

A short piece of DNA that pairs with the sequence on one side of a target and gives a polymerase its starting point; PCR uses two, one on either side.

Principal component analysis (PCA)

A method that finds the directions of greatest variation among samples, used to spot batches and outliers.

Protein group

The set of proteins that the same identified peptides could have come from, reported in place of a single protein when peptides are shared.

Protein inference

Deducing which proteins were present from the peptides identified; ambiguous when peptides are shared.

Proteoform

A specific molecular form of a protein, including its sequence variant, splicing and modifications.

Proteome

The full set of proteins in a cell or sample, recording which proteins have been made and in what modified forms.

Proximity extension assay

An affinity method in which two DNA-tagged antibodies must bind the same protein for their tags to form a countable barcode.

Quality Management System Regulation (QMSR)

FDA's rule for device makers' quality systems, 21 CFR Part 820, in force since 2 February 2026; it incorporates ISO 13485:2016 by reference.

Quality score (Q)

A logarithmic estimate of a base call's error probability: Q30 is one error in 1,000.

Read

The sequence that an instrument reports for one piece of DNA.

Recall (India)

Under India's Medical Devices Rules, any action by the maker, its authorized agent or a supplier to remove a device from the market or retrieve it from those it was supplied to, because it is hazardous to health, fails to conform to a claim about its quality, safety or efficacy, or does not meet the Act and the Rules.

Reclassification order

An FDA order that moves a type of device from one class to another, first proposed and then made final; a proposed order changes nothing until it is final.

Reference genome

An agreed genome sequence used as the coordinate system for aligning reads.

Reference material

A material whose answer is known in advance. It tests the links of a test's chain from the point where it enters, and only where its values are known, because the difference between the known answer and the report measures the error of those links.

Relative abundance

An organism's share of the reads in a community sample; the shares sum to 100 %, so a rise in one organism lowers the shares of all the others.

Removal (21 CFR 806)

Moving a device from its point of use to another location.

Research Use Only (RUO)

In the US, the labeling statement "For Research Use Only. Not for use in diagnostic procedures." for a product in the laboratory research phase. In India, CDSCO treats research-use products as outside the Medical Devices Rules provided they are not intended for diagnosis.

Residual risk

The risk remaining after risk control measures have been implemented (ISO 14971, as FDA reproduces it), judged again against the criteria of the risk management plan.

Retention time

The time a molecule takes to emerge from a chromatography column, used as an identity coordinate.

Reverse transcriptase

An enzyme that copies RNA into DNA.

Risk (ISO 14971)

The combination of the probability of occurrence of harm and the severity of that harm.

Risk management plan

The document that sets out the scope, responsibilities and criteria of risk management, so that the criteria for accepting a risk are fixed before any risk is estimated.

RNA-seq

RNA sequencing: measuring the levels of RNA molecules by converting them to cDNA, sequencing it and counting the reads that land on each gene.

SAM and BAM

File formats for reads aligned to a reference: SAM is tab-delimited text, BAM the same data compressed.

Sample index

A short known sequence in a library's adapters that lets many libraries share one flow cell and be sorted back by sample.

Sandwich immunoassay

An assay, often called an ELISA, in which one antibody fixed to a surface captures a protein and a second antibody, against a different part of it, binds on top and carries an enzyme that amplifies the signal.

Scientific validity (IVDR)

The association of an analyte with a clinical condition or a physiological state.

Segmentation

The step in spatial methods that decides which molecules belong to which cell, usually by drawing cell boundaries from a nuclear or membrane stain.

Sensitivity

The share of true cases that a test detects, such as the share of people with a condition whom the test calls positive; also called diagnostic sensitivity, to set it apart from analytical sensitivity.

Sequencing depth (RNA-seq)

The total number of reads a sample receives; genome laboratories use depth for the coverage of each position.

Serious incident (IVDR)

An incident that directly or indirectly led, might have led or might lead to a death, a temporary or permanent serious deterioration in a person's state of health, or a serious public health threat. The maker reports it within 15 days of becoming aware, 10 days for a death or an unanticipated serious deterioration, and 2 days for a serious public health threat.

Service laboratory

A company laboratory that sequences or measures samples for clients and returns data or reports, rather than selling the instruments or kits.

Shotgun metagenomics

Sequencing all the DNA in a mixed sample, without amplifying a marker gene, so that reads are assigned to species and strains and the community's genes are cataloged.

Side scatter

Light scattered by a cell to the side, which grows with its internal complexity, such as its granules and the shape of its nucleus.

Signature

A pattern across many measurements that separates one group of people or samples from another.

Simplex mode (sequencing by expansion)

The sequencing-by-expansion mode that Roche describes as built for throughput, for which it states reads of up to 1,500 bases under suitable sample and library conditions.

Single-nucleotide polymorphism (SNP)

A position in the genome at which people commonly carry one of two different letters; a panel of them read by two independent assays exposes a sample swap.

Software of unknown provenance (SOUP)

In IEC 62304, software already developed and generally available but not developed for incorporation into the device, or software developed earlier for which adequate records of its development are not available.

Sparse (single-cell data)

Of data in which most genes read zero in any given cell, because only a fraction of each cell's RNA is captured and converted.

Spatial methods

Methods that keep each measurement's position in a tissue, either by barcoded capture arrays or by imaging molecules in place.

Special controls

Requirements the FDA writes for one type of class II device, in addition to the general controls.

Specificity

The share of samples without the target, or of people without the condition, that a test correctly reports as negative.

Spectral cytometry

Cytometry that records each cell's full emission spectrum and unmixes the dyes computationally.

Spike-in

A known quantity of a reference organism, molecule or DNA added to every sample, to test and calibrate the measurement chain or to turn relative amounts into absolute ones.

Spillover

The light from one fluorescent dye caught by a detector meant for another, because each dye emits across a broad band of wavelengths.

State (of a cell)

What a cell is at a given moment: which of its genome's instructions it is using, how intensely and with what result.

State Licensing Authority

Under India's Medical Devices Rules, the authority of an Indian state that licenses the manufacture of class A and B devices made there.

Structural variant

A genomic change of more than about 50 base pairs: a deletion, duplication, inversion or insertion.

Supplier

A company from which a device maker obtains a part, material, software component or service that goes into the device. Under ISO 13485, which the QMSR incorporates, the criteria for evaluating and selecting suppliers are proportionate to the risk associated with the device.

Suspected unexpected serious adverse event (India)

The event that a license holder under India's Medical Devices Rules reports to its licensing authority, with any action taken, including a recall, within 15 days of learning of it (Rule 26(ii)).

Targeted metabolomics

The measurement of a predefined set of metabolites, each with an isotope-labeled internal standard, reported as absolute concentrations.

Technical positive predictive value (TPPV)

The share of a test's variant calls that a reference confirms.

Test (as a chain)

The whole chain from the patient's specimen to a signed report: specimen, pre-analytical handling, library or assay preparation, instrument and primary analysis, pipeline, interpretation and report.

Third-party review (510(k))

A route in which an organization accredited by the FDA reviews a 510(k) submission before the FDA makes the decision.

Top-down proteomics

Mass spectrometry that weighs and fragments intact proteins, measuring proteoforms whole, at lower throughput than bottom-up work.

Traceability

The links from each requirement to the outputs that meet it and the records that prove it.

Trajectory

An ordering of single cells caught at different stages of a continuous process, such as differentiation.

Transcript length

The length of an RNA molecule, which affects its count, because a longer transcript is cut into more fragments and collects more reads at the same number of molecules.

Transcriptome

The complete set of RNA molecules in a cell or sample at a given time.

Trend reporting (IVDR)

The maker's watch, under Article 83, over non-serious incidents and expected erroneous results for a statistically significant increase in their frequency or severity; MDCG guidance asks the maker to define the methods, thresholds and observation period in its post-market surveillance plan.

Tumor genome profiling

Reading tens to hundreds of genes in a biopsy to find mutations that match a targeted drug.

UKCA marking

The marking under which a device is sold in Great Britain.

Unique dual indexes

A pair of indexes, one at each end of a library, neither shared with any other library in the pool, so that a read whose indexes disagree can be set aside.

Unique molecular identifier (UMI)

A random tag attached to each captured molecule so that amplification copies are counted once.

Untargeted (metabolomics)

Of a run that records whatever ionizes rather than a chosen list of compounds, detecting features.

User facility (US reporting)

In FDA's reporting rules, a hospital, ambulatory surgical facility, nursing home or outpatient diagnostic or treatment facility; it reports device-related deaths to the FDA and the maker, and serious injuries to the maker, within 10 work days.

Validation (design)

Evidence, on representative product and before release, that the finished design meets its users' needs and intended use.

Variant calling

Identifying positions where the reads consistently differ from the reference, with a confidence for each call.

VCF (variant call format)

The file format for variant calls, with one data line per genomic position and one column per sample.

Verification (design)

Evidence that each design output meets its design input.

Volcano plot

A plot of every feature by its log₂ fold change across the page and by its significance, in tenfold steps of the p-value, up the page.

WHO prequalification (IVDs)

WHO's assessment of in vitro diagnostics for listed analytes. Its eligibility tables, last updated on 27 September 2026, list what may be assessed, and their only sequencing entry is targeted sequencing for drug resistance in tuberculosis.

Work day (US reporting)

In FDA's rules for medical device reporting, Monday through Friday except federal holidays; with no holiday, 5 work days from any weekday end on the seventh calendar day.

y ions

The fragments of a peptide that contain its end; they form a ladder whose rungs differ by the mass of one amino acid, so the gaps read the sequence from the end.

Sources are listed by the chapter in which they are first used. A note after a citation says how it was read where that bears on its weight, for example from the abstract only, through a secondary report or in the author manuscript, and marks a manufacturer's document or a company's statement as such. Figures from manufacturers are also identified as such in the text. Status and prices are as of October 2026.

1. Opening — Cranford N. NASA Twins Study Confirms Preliminary Findings. NASA, 31 January 2018, with a statement added by NASA on 15 March 2018 (page last updated 12 October 2023). https://www.nasa.gov/humans-in-space/nasa-twins-study-confirms-preliminary-findings/ (the study's sponsor; preliminary results, superseded by the 2019 paper)
2. Opening — NASA. Veteran Station Crew Returns to Earth after Historic Mission. Space Station blog, 1 March 2016. https://www.nasa.gov/blogs/spacestation/2016/03/01/veteran-station-crew-returns-to-earth-after-historic-mission/ (mission record)
3. Opening — CNN Wire. Astronaut's DNA no longer matches identical twin's after time in space, NASA finds. Syndicated wire story, as published by WTVR CBS 6 Richmond, 14 March 2018. https://wtvr.com/2018/03/14/astronauts-dna-no-longer-matches-identical-twins-after-time-in-space-nasa-finds (no individual byline; no correction shown on the copy read, 24 October 2026)
4. Opening — Drake N. No, Scott Kelly's Year in Space Didn't Mutate His DNA. National Geographic, 15 March 2018. https://www.nationalgeographic.com/science/article/scott-kelly-astronaut-space-station-dna-health-science (news report; the source of the quotations from Christopher Mason and Susan Bailey)
5. Opening — NASA. NASA's Twins Study Results Published in Science Journal. Feature, 11 April 2019, https://www.nasa.gov/feature/nasa-s-twins-study-results-published-in-science ; Twins Study: Research, web page, https://www.nasa.gov/twins-study/research (the study's sponsor; page last updated 12 October 2023; the count of ten research teams is NASA's and does not appear in the 2019 paper)
6. Opening — Garrett-Bakelman F.E., Darshi M., Green S.J., … Snyder M.P., Turek F.W. The NASA Twins Study: a multidimensional analysis of a year-long human spaceflight. Science 364(6436), eaau8650 (2019), published 12 April 2019. doi:10.1126/science.aau8650 (read in the author manuscript; 82 authors, the first at Weill Cornell Medicine and the University of Virginia; funded by NASA, NIH, NSF, DLR and private foundations; five authors declared roles in companies as founders, board members, advisers or consultants)
7. Ch. 1 — BioNumbers. Counts of cell types in the human body, BNID 103626 (200, from Junqueira et al., Basic Histology, 7th ed., 1992) and BNID 106155 (411, from Vickaryous and Hall 2006). Database records. https://bionumbers.hms.harvard.edu/bionumber.aspx?id=103626 ; https://bionumbers.hms.harvard.edu/bionumber.aspx?id=106155 (secondary compilation; accessed 24 October 2026)
8. Ch. 1 — Vickaryous M.K., Hall B.K. Human cell type diversity, evolution, development, and classification with special reference to cells derived from the neural crest. Biological Reviews 81, 425–455 (2006). PMID 16790079 (read through BioNumbers; the review itself not opened)
9. Ch. 1 — Ensembl. Homo sapiens annotation page, release 116 (GRCh38.p14 golden path length 3,099,750,718 bp). June 2026. https://jun2026.archive.ensembl.org/Homo_sapiens/Info/Annotation
10. Ch. 1 — GENCODE. Human statistics, release 50 (19,442 protein-coding genes). Web page (read 24 October 2026). https://www.gencodegenes.org/human/stats.html
11. Ch. 1 — BioNumbers. BNID 110283: protein-coding sequence is "less than 1.5%" of the human genome, citing International Human Genome Sequencing Consortium, Initial sequencing and analysis of the human genome, Nature 409, 860–921 (2001). Database record. https://bionumbers.hms.harvard.edu/bionumber.aspx?id=110283 (secondary compilation; the 2001 paper not opened)
12. Ch. 1 — Gurdon J.B. The developmental capacity of nuclei taken from intestinal epithelium cells of feeding tadpoles. Journal of Embryology and Experimental Morphology 10(4), 622–640 (1962). Copy read: https://www.nig.ac.jp/jimu/soken/courses/devbiol/2025/develop_10_4_622.pdf
13. Ch. 1 — Gurdon J.B. The cloning of a frog. Development 140(12), 2446–2448 (2013). doi:10.1242/dev.097899 (the author's own retrospective: the work begun as a graduate student of Michail Fischberg, in Xenopus laevis)
14. Ch. 1 — Schwanhäusser B. … Selbach M. Global quantification of mammalian gene expression control. Nature 473, 337–342 (2011). doi:10.1038/nature10098 (read in the Max Delbrück Center repository manuscript; one author at MicroDiscovery GmbH; the paper was corrected on 13 February 2013)
15. Ch. 2 — Said E.A. et al. Defining IL-6 levels in healthy individuals: a meta-analysis. Journal of Medical Virology 93, 3915–3924 (2021; online 22 November 2020). doi:10.1002/jmv.26654 (no funder listed in the Crossref record)
16. Ch. 2 — National Human Genome Research Institute. DNA Sequencing Costs: Data. Web page, last updated 16 May 2023. https://www.genome.gov/about-genomics/fact-sheets/DNA-Sequencing-Costs-Data (production costs at the sequencing centers the institute funds: labor, reagents, instruments amortized over three years and indirect costs included; quality assessment, technology development, project management and data analysis beyond the first processing excluded)
17. Ch. 2 — Anderson N.L., Anderson N.G. The human plasma proteome: history, character, and diagnostic prospects. Molecular & Cellular Proteomics 1(11), 845–867 (2002). doi:10.1074/mcp.R200007-MCP200 (abstract only)
18. Ch. 2 — UniProt Consortium. UniProtKB entries P02768 (serum albumin, human) and P05231 (interleukin-6, human). Database records, accessed 24 October 2026 (the molecular weights of the solved example). https://rest.uniprot.org/uniprotkb/search?query=accession:P02768+OR+accession:P05231&fields=accession,protein_name,length,mass,ft_chain,ft_signal,ft_propep&format=tsv
19. Ch. 3 — Thermo Fisher Scientific. Orbitrap Exploris 240 mass spectrometer. Product specifications PS73409-EN (0626S). https://documents.thermofisher.com/TFS-Assets/CMD/Specification-Sheets/ps-73409-ms-orbitrap-exploris-240-ps73409-en.pdf (manufacturer; mass accuracy under the maker's defined conditions)
20. Ch. 3 — Stoeckius M., Hafemeister C., Stephenson W., … Smibert P. Simultaneous epitope and transcriptome measurement in single cells. Nature Methods 14(9), 865–868 (2017). doi:10.1038/nmeth.4380 (bibliographic record only; New York Genome Center)
21. Ch. 4 — Rhie A. et al. The complete sequence of a human Y chromosome. Nature 621, 344–354 (2023). doi:10.1038/s41586-023-06457-y (competing interests: one author an employee of Oxford Nanopore Technologies; other authors with travel funding, research support or patents involving instrument makers; funded by NHGRI, other NIH institutes and HHMI)
22. Ch. 4 — Illumina, Inc. Quality Scores for Next-Generation Sequencing. Web page, undated (manufacturer; the Phred scale, citing Ewing and Green, Genome Research 8, 1998). https://www.illumina.com/science/technology/next-generation-sequencing/plan-experiments/quality-scores.html
23. Ch. 4 — Byrska-Bishop M. et al. High-coverage whole-genome sequencing of the expanded 1000 Genomes Project cohort including 602 trios. Cell 185(18), 3426–3440.e19 (2022). doi:10.1016/j.cell.2022.08.004 (abstract only; New York Genome Center)
24. Ch. 4 — Nurk S. … Phillippy A.M. The complete sequence of a human genome. Science 376(6588), 44–53 (2022). doi:10.1126/science.abj6987 (abstract only; 100 authors of the Telomere-to-Telomere consortium, among them staff of Pacific Biosciences)
25. Ch. 4 — Liao W.-W. … Paten B. A draft human pangenome reference. Nature 617(7960), 312–324 (2023). doi:10.1038/s41586-023-05896-x (abstract only; Human Pangenome Reference Consortium)
26. Ch. 4 — Human Pangenome Reference Consortium. Announcing the Human Pangenome Reference Consortium Data Release 2. Web page, 12 May 2025. https://humanpangenome.org/news/hprc-data-release-2 (a data release, not a peer-reviewed paper)
27. Ch. 4 — Bhattacharyya C., Subramanian K. et al. (GenomeIndia Consortium). Mapping genetic diversity with the GenomeIndia project. Nature Genetics 57(4), 767–773 (2025). doi:10.1038/s41588-025-02153-x (abstract only; funded by the Department of Biotechnology, Government of India; an author correction of 8 May 2025, doi:10.1038/s41588-025-02206-1, amends author names only)
28. Ch. 4 — Press Information Bureau, Government of India (Ministry of Science and Technology). India Takes a Giant Leap in Genomics: Launch of Indian Genomic Data Set & IBDC Portals to Empower Global Research. Press release, 9 January 2025. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2091577&reg=48&lang=2
29. Ch. 4 — Mahmoud M. et al. Structural variant calling: the long and the short of it. Genome Biology 20, 246 (2019). doi:10.1186/s13059-019-1828-7
30. Ch. 4 — Illumina, Inc. Whole-exome sequencing. Web page, undated (manufacturer; the share of known disease-related variants that lie in the exome, citing van Dijk et al., Trends in Genetics 30, 2014, not opened). https://www.illumina.com/techniques/sequencing/dna-sequencing/targeted-resequencing/exome-sequencing.html
31. Ch. 5 — Sanger F., Nicklen S., Coulson A.R. DNA sequencing with chain-terminating inhibitors. Proceedings of the National Academy of Sciences USA 74(12), 5463–5467 (1977). doi:10.1073/pnas.74.12.5463 (abstract only)
32. Ch. 5 — Zawilak T. Sanger Sequencing. Embryo Project Encyclopedia, Arizona State University, 8 August 2025. https://embryo.asu.edu/pages/sanger-sequencing (secondary)
33. Ch. 5 — Thermo Fisher Scientific. Sanger Sequencing by CE 1: Foundations. Behind the Bench blog, undated. https://www.thermofisher.com/blog/behindthebench/sanger-sequencing-by-ce-1-foundations/ (manufacturer)
34. Ch. 5 — National Human Genome Research Institute. Human Genome Project. Fact sheet. https://www.genome.gov/about-genomics/educational-resources/fact-sheets/human-genome-project
35. Ch. 5 — Illumina, Inc. NovaSeq X Series Specifications. Web page, undated (manufacturer). https://www.illumina.com/systems/sequencing-platforms/novaseq-x-plus/specifications.html ; Transition to High-Throughput with the NovaSeq X Series. Web page, undated (manufacturer; the "$200 genome" at list price on the 25B flow cell, coverage not stated). https://www.illumina.com/systems/sequencing-platforms/novaseq-x-plus/applications/transition.html
36. Ch. 5 — Pacific Biosciences of California, Inc. PacBio SPRQ-Nx Chemistry Now Shipping Worldwide, Enabling Sub-$300 HiFi Genomes for Large-Scale Projects and AI-Enhanced Sequencing. Press release. 26 May 2026 (company statement; prices for sequencing reagents only, at list price, the lower figure at 5,000 genomes a year with discounts); HiFi Reads web page, last modified 19 March 2026 (manufacturer); Revio web page, undated (manufacturer; HiFi reads typically 15 to 20 kb). https://www.pacb.com/press_releases/pacbio-sprq-nx-chemistry-now-shipping-worldwide-enabling-sub-300-hifi-genomes-for-large-scale-projects-and-ai-enhanced-sequencing/ ; https://www.pacb.com/technology/hifi-sequencing/ ; https://www.pacb.com/revio/
37. Ch. 5 — Oxford Nanopore Technologies. Nanopore sequencing accuracy. Web page, undated (manufacturer; raw read accuracy with Dorado v5 basecalling models; methylation calls on a synthetic truth set); From kilobases to 'whales': a short history of ultra-long reads and high-throughput genome sequencing. Blog, 5 July 2021 (manufacturer). https://nanoporetech.com/platform/accuracy ; https://nanoporetech.com/blog/news-blog-kilobases-whales-short-history-ultra-long-reads-and-high-throughput-genome
38. Ch. 5 — F. Hoffmann-La Roche Ltd. Roche announces the launch of AXELIOS 1, a transformative next-generation sequencing platform. Media release, Basel. 29 June 2026 (company statement; research use only); Roche Diagnostics. Overview of sequencing by expansion (SBX) technology. Web page (manufacturer; read in part). https://www.roche.com/media/releases/med-cor-2026-06-29 ; https://diagnostics.roche.com/global/en/diagnostics-insights/sbx-technology-overview.html
39. Ch. 5 — Our World in Data. Cost of sequencing a full human genome. Data file (CSV), last updated 28 November 2023 (secondary: the National Human Genome Research Institute's series with minor processing, the source of the 2001 and 2022 values). https://ourworldindata.org/grapher/cost-of-sequencing-a-full-human-genome.csv
40. Ch. 5 — Ultima Genomics, Inc. Ultima Genomics increases output by over 50% and lowers sequencing costs by a further 20% with the introduction of UG 100 Solaris. Press release (PR Newswire). 24 February 2025 (company statement: "enabling the $80 genome", without its coverage or what it includes). https://www.prnewswire.com/news-releases/ultima-genomics-increases-output-by-over-50-and-lowers-sequencing-costs-by-a-further-20-with-the-introduction-of-ug-100-solaris-302382977.html
41. Ch. 6 — Horvath S. DNA methylation age of human tissues and cell types. Genome Biology 14, R115 (2013). doi:10.1186/gb-2013-14-10-r115 (no specific funding; competing interests: a provisional patent application owned by the Regents of the University of California, with Horvath as inventor)
42. Ch. 6 — Lövkvist C., Dodd I.B., Sneppen K., Haerter J.O. DNA methylation in human epigenomes depends on local topology of CpG sites. Nucleic Acids Research 44, 5123–5132 (2016). doi:10.1093/nar/gkw124 (the count of 28 million CpG sites, on the hg18 build)
43. Ch. 6 — ENCODE Consortium. WGBS data standards. Web page, updated June 2017. https://www.encodeproject.org/data-standards/wgbs/
44. Ch. 6 — Lister R. … Ecker J.R. Global epigenomic reconfiguration during mammalian brain development. Science 341(6146), 1237905 (2013). doi:10.1126/science.1237905 (abstract only)
45. Ch. 6 — Vaisvila R. … Davis T.B. Enzymatic methyl sequencing detects DNA methylation at single-base resolution from picograms of DNA. Genome Research 31(7), 1280–1289 (2021). doi:10.1101/gr.266551.120 (abstract only; all 20 authors at New England Biolabs, the maker of the enzymatic kit, which funded the study)
46. Ch. 6 — Huang Y., Pastor W.A., Shen Y., Tahiliani M., Liu D.R., Rao A. The behaviour of 5-hydroxymethylcytosine in bisulfite sequencing. PLoS ONE 5(1), e8888 (2010). doi:10.1371/journal.pone.0008888 (abstract only)
47. Ch. 6 — Buenrostro J.D., Giresi P.G., Zaba L.C., Chang H.Y., Greenleaf W.J. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nature Methods 10(12), 1213–1218 (2013). doi:10.1038/nmeth.2688 (bibliographic record only; Stanford University)
48. Ch. 6 — Kaya-Okur H.S., Wu S.J., Codomo C.A., Pledger E.S., Bryson T.D., Henikoff J.G., Ahmad K., Henikoff S. CUT&Tag for efficient epigenomic profiling of small samples and single cells. Nature Communications 10, 1930 (2019). doi:10.1038/s41467-019-09982-5 (Fred Hutchinson Cancer Research Center; the authors declared no competing interests)
49. Ch. 6 — Lieberman-Aiden E. … Dekker J. Comprehensive mapping of long-range interactions reveals folding principles of the human genome. Science 326(5950), 289–293 (2009). doi:10.1126/science.1181369 (abstract only; Broad Institute and University of Massachusetts Medical School)
50. Ch. 7 — Nagaraj N. et al. Deep proteome and transcriptome mapping of a human cancer cell line. Molecular Systems Biology 7, 548 (2011). doi:10.1038/msb.2011.81 (read through BioNumbers BNID 108426: at least 10,000 to 12,000 genes expressed in HeLa cells)
51. Ch. 7 — Al'Khafaji A.M. et al. High-throughput RNA isoform sequencing using programmed cDNA concatenation. Nature Biotechnology 42, 582–586 (2024; online 8 June 2023). doi:10.1038/s41587-023-01815-7 (competing interests: a licensed, pending international patent application; a collaboration agreement between the Broad Institute and Pacific Biosciences entered after submission)
52. Ch. 7 — ENCODE Consortium. RNA-seq data standards (long RNAs). Web page, updated June 2017. https://www.encodeproject.org/data-standards/rna-seq/long-rnas/
53. Ch. 7 — Milo R., Phillips R. How many mRNAs are in a cell? Cell Biology by the Numbers, online book, undated. https://book.bionumbers.org/how-many-mrnas-are-in-a-cell/ ; BioNumbers BNID 109916, https://bionumbers.hms.harvard.edu/bionumber.aspx?id=109916 (secondary compilation: about 200,000 messenger RNA molecules in a typical mammalian cell, citing Shapiro, Biezuner and Linnarsson 2013, not opened)
54. Ch. 7 — Pan Q. … Blencowe B.J. Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing. Nature Genetics 40(12), 1413–1415 (2008). doi:10.1038/ng.259 (abstract only)
55. Ch. 7 — Paik S. … Wolmark N. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. New England Journal of Medicine 351(27), 2817–2826 (2004). doi:10.1056/NEJMoa041588 (abstract only)
56. Ch. 8 — Beimers W.F., Overmyer K.A., Sinitcyn P., Lancaster N.M., Quarmby S.T., Coon J.J. Technical Evaluation of Plasma Proteomics Technologies. Journal of Proteome Research 24(6), 3074–3087 (2025). doi:10.1021/acs.jproteome.5c00221 (abstract only; the study was initiated by Seer, which paid for some of its assays and the plasma, per the trade report below)
57. Ch. 8 — Beimers W.F. et al. A Technical Evaluation of Plasma Proteomics Technologies. bioRxiv 2025.01.08.632035, posted 13 January 2025. doi:10.1101/2025.01.08.632035 (preprint of the paper above; the source of the figure of about 600 proteins for neat plasma)
58. Ch. 8 — GenomeWeb. Benchmarking study evaluates performance tradeoffs of plasma proteomic workflows. News report, January 2025. https://www.genomeweb.com/proteomics-protein-research/benchmarking-study-evaluates-performance-tradeoffs-plasma-proteomic (trade press; Seer's role in the study and the senior author's links to Seer and Thermo Fisher Scientific)
59. Ch. 8 — HMDB. Metabocard HMDB0000122, D-Glucose (formula C6H12O6, monoisotopic mass 180.0634; entry updated 30 May 2023). https://hmdb.ca/metabolites/HMDB0000122 (glucose read; PubChem's record for the three sugars could not be opened)
60. Ch. 8 — Deutsch E.W., Lindskog C., Budnik B., Ctortecka C., Guo T. et al. The 2025 Report on the Human Proteome from the HUPO Human Proteome Project. Journal of Proteome Research 25(2), 539–555 (2026). doi:10.1021/acs.jproteome.5c00759 (abstract only)
61. Ch. 8 — Guzman U.H., Martinez-Val A., Ye Z., Damoc E., Arrey T.N. … Makarov A., Zabrouskov V., Olsen J.V. Ultra-fast label-free quantification and comprehensive proteome coverage with narrow-window data-independent acquisition. Nature Biotechnology 42, 1855–1866 (2024). doi:10.1038/s41587-023-02099-7 (University of Copenhagen; 14 of the 22 authors at Thermo Fisher Scientific, the instrument's maker)
62. Ch. 8 — UniProt Consortium. UniProtKB P02769 (ALBU_BOVIN), albumin, Bos taurus. Database record, accessed 24 October 2026. https://rest.uniprot.org/uniprotkb/P02769.fasta
63. Ch. 8 — Matrix Science Ltd. Mascot database search: amino acid reference data. Web page (residue masses). https://www.matrixscience.com/help/aa_help.html
64. Ch. 8 — Völlmy F., Kaushik P., Eckert S., Paschen C., Tavalaei S., Fidelin J., Weiser S., Bantscheff M. Comparison of Common Plasma Proteomics Workflows Reveals Distinct Preanalytical Biases by Sample Preparation and Nanoparticle Enrichment. Journal of Proteome Research 25(6), 3178–3187 (2026). doi:10.1021/acs.jproteome.6c00048 (abstract only; all authors at Roche, which funded the study)
65. Ch. 9 — Assarsson E., Lundberg M., Holmquist G. et al. Homogenous 96-Plex PEA Immunoassay Exhibiting High Sensitivity, Specificity, and Excellent Scalability. PLoS ONE 9(4), e95192 (2014). doi:10.1371/journal.pone.0095192 (all authors but two employees of Olink, the company commercializing the method)
66. Ch. 9 — Olink (part of Thermo Fisher Scientific). Olink Explore HT. Product page, undated (manufacturer: about 5,400 proteins from 2 µL of plasma, read by sequencing; thousands of samples a week). https://olink.com/products/olink-explore-ht
67. Ch. 9 — Feng W. et al. NULISA: a proteomic liquid biopsy platform with attomolar sensitivity and high multiplexing. Nature Communications 14, 7238 (2023). doi:10.1038/s41467-023-42834-x (authors at Alamar Biosciences, the method's developer, and the University of Bonn)
68. Ch. 9 — Gold L., Ayers D., Bertino J. et al. Aptamer-Based Multiplexed Proteomic Technology for Biomarker Discovery. PLoS ONE 5(12), e15004 (2010). doi:10.1371/journal.pone.0015004 (most authors employees of SomaLogic, which funded the work)
69. Ch. 9 — SomaLogic, Inc. SomaLogic Announces New SomaScan 11K Platform. Press release (GlobeNewswire, via TipRanks), Boulder. 1 November 2023 (company statement). https://www.tipranks.com/news/press-releases/somalogic-announces-new-somascan-11k-platform
70. Ch. 9 — Eldjarn G.H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature 622, 348–358 (2023). doi:10.1038/s41586-023-06563-x (abstract only; all 33 authors at deCODE Genetics or Amgen)
71. Ch. 9 — Sun B.B. et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature 622, 329–338 (2023). doi:10.1038/s41586-023-06592-6 (49 of the 56 named authors list a company; the measurements were funded by the UK Biobank Pharma Proteomics Project, a consortium of 13 biopharmaceutical companies)
72. Ch. 9 — UK Biobank. Launch of world's most significant protein study. News, January 2025. https://www.ukbiobank.ac.uk/?p=19134 (funded by a consortium of 14 biopharmaceutical companies)
73. Ch. 9 — Quantum-Si Inc. Quantum-Si Begins Commercial Shipments of the Platinum Protein Sequencing System and Provides Business Updates at the 41st Annual J.P. Morgan Healthcare Conference. Press release (Business Wire). 9 January 2023 (company statement). https://www.businesswire.com/news/home/20230109005274/en
74. Ch. 9 — Quantum-Si Inc. Quantum-Si Launches Platinum Pro: The End-to-End Solution for Deeper Insights in Proteomics. Press release (Business Wire). 13 January 2025 (company statement). https://www.businesswire.com/news/home/20250113845303/en/Quantum-Si-Launches-Platinum-Pro-The-End-to-End-Solution-for-Deeper-Insights-in-Proteomics
75. Ch. 9 — Reed B.D., Meyer M.J., Abramzon V. et al. Real-time dynamic single-molecule protein sequencing on an integrated semiconductor device. Science 378(6616), 186–192 (2022). doi:10.1126/science.abo7651 (abstract only; 73 of the 80 authors at Quantum-Si)
76. Ch. 9 — Quantum-Si Inc. Quantum-Si Announces Detection of 18 Amino Acids Using a Developmental Sequencing Kit on Proteus. Press release (GlobeNewswire, via Finviz), Branford, Connecticut. 15 September 2026 (company statement). https://finviz.com/news/391959/quantum-si-announces-detection-of-18-amino-acids-using-a-developmental-sequencing-kit-on-proteus
77. Ch. 9 — Quantum-Si Inc. Quantum-Si Demonstrates Detection of 19 Amino Acids in Proteus Development Program. Press release (GlobeNewswire, via Finviz). 8 October 2026 (company statement; commercial launch of Proteus expected in the second quarter of 2027). https://finviz.com/news/400600/quantum-si-demonstrates-detection-of-19-amino-acids-in-proteus-development-program
78. Ch. 9 — Motone K. … Nivala J. Multi-pass, single-molecule nanopore reading of long protein strands. Nature 633(8030), 662–669 (2024). doi:10.1038/s41586-024-07935-7 (University of Washington; the strands were read on Oxford Nanopore flow cells, and the paper's bibliographic record lists Oxford Nanopore Technologies among the funders)
79. Ch. 9 — Nautilus Biotechnology, Inc. Nautilus Biotechnology Launches Early Access Program for Novel Single-Molecule Proteomics Platform, Starting with Tau Proteoforms Assay. Press release (GlobeNewswire, via BioSpace), Seattle. 8 January 2026; Nautilus Biotechnology Unveils Voyager Platform, Enabling Single-Molecule Iterative Mapping, at US HUPO 2026. Press release (via Nasdaq). 24 February 2026 (company statements). https://www.biospace.com/press-releases/nautilus-biotechnology-launches-early-access-program-for-novel-single-molecule-proteomics-platform-starting-with-tau-proteoforms-assay ; https://www.nasdaq.com/press-release/nautilus-biotechnology-unveils-voyager-platform-enabling-single-molecule-iterative
80. Ch. 9 — US FDA. Testing for Detection and Differentiation of Influenza A Virus Subtypes Using Multiplex Assays: Class II Special Controls Guidance for Industry and FDA Staff. 9 October 2009. https://www.fda.gov/media/77204/download
81. Ch. 10 — Wishart D.S. … Gautam V. HMDB 5.0: the Human Metabolome Database for 2022. Nucleic Acids Research 50(D1), D622–D631 (2022). doi:10.1093/nar/gkab1062 (abstract only); HMDB About and Statistics pages, accessed 24 October 2026 (the counts of the solved example). https://hmdb.ca/about ; https://hmdb.ca/statistics
82. Ch. 10 — Emwas A.-H., Roy R., McKay R.T. … Wishart D.S. NMR Spectroscopy for Metabolomics Research. Metabolites 9(7), 123 (2019). doi:10.3390/metabo9070123
83. Ch. 10 — Chi Y., Mitchell J.M., Zheng S., Thapa M., Li S. Assessing the metabolomics "dark matter" by a detectable khipu model. Metabolomics 22(3), 79 (2026; online 24 May 2026). doi:10.1007/s11306-026-02456-y; preprint bioRxiv doi:10.1101/2025.02.04.636472, posted 5 February 2025 (The Jackson Laboratory for Genomic Medicine)
84. Ch. 10 — Sumner L.W., Amberg A., Barrett D. … Viant M.R. Proposed minimum reporting standards for chemical analysis. Chemical Analysis Working Group (CAWG), Metabolomics Standards Initiative (MSI). Metabolomics 3(3), 211–221 (2007). doi:10.1007/s11306-007-0082-2
85. Ch. 10 — CDC. Using Tandem Mass Spectrometry for Metabolic Disease Screening Among Newborns: A Report of a Work Group. MMWR Recommendations and Reports 50(RR-3), 13 April 2001. https://www.cdc.gov/mmwr/preview/mmwrhtml/rr5003a1.htm
86. Ch. 10 — Health Resources and Services Administration. Recommended Uniform Screening Panel. Web page, tables as of January 2023, page last reviewed July 2024. https://www.hrsa.gov/advisory-committees/heritable-disorders/rusp
87. Ch. 10 — LIPID MAPS. LIPID MAPS Structure Database (LMSD) overview. Web page. https://www.lipidmaps.org/databases/lmsd/overview
88. Ch. 10 — Reusch D., Tejada M.L. Fc glycans of therapeutic antibodies as critical quality attributes. Glycobiology 25(12), 1325–1334 (2015). doi:10.1093/glycob/cwv065 (abstract only; the authors at Roche Diagnostics and Genentech)
89. Ch. 11 — Konecny A.J., Mage P.L., Tyznik A.J., Prlic M., Mair F. OMIP-102: 50-color phenotyping of the human immune system with in-depth assessment of T cells and dendritic cells. Cytometry Part A 105(6), 430–436 (2024). doi:10.1002/cyto.a.24841 (abstract only; Mage and Tyznik at BD Biosciences, which made the instrument)
90. Ch. 11 — Cytek Biosciences, Inc. Cytek Biosciences to Unveil Next-Generation Spectral Flow Cytometry Innovations at CYTO 2026. Press release, Fremont. 2 June 2026 (company statement: the Borealis spectral cytometer, its 60-color panels and an early-access program). https://investors.cytekbio.com/news-releases/news-release-details/cytek-biosciences-unveil-next-generation-spectral-flow-cytometry
91. Ch. 11 — Iyer A., Hamers A.A.J., Pillai A.B. CyTOF for the Masses. Frontiers in Immunology 13, 815828 (2022). doi:10.3389/fimmu.2022.815828 (review; University of Miami)
92. Ch. 12 — Physicians Committee for Responsible Medicine. Human Cell Atlas consortium shares latest collection of foundational studies. News, 6 January 2025 (secondary: the data portal's count of about 62 million cells from about 9,100 donors, citing the Nature collection of 20 November 2024). https://www.pcrm.org/news/innovative-science/human-cell-atlas-consortium-shares-latest-collection-foundational-studies
93. Ch. 12 — 10x Genomics, Inc. 10x Genomics Delivers Single Cell for a Single Cent with New Chromium Launches. Press release. 15 October 2024 (company statement). https://investors.10xgenomics.com/news/news-details/2024/10x-Genomics-Delivers-Single-Cell-for-a-Single-Cent-with-New-Chromium-Launches/default.aspx
94. Ch. 12 — Parse Biosciences. Evercode WT Mega. Product page, undated (manufacturer: a million or more cells per kit, without droplets). https://www.parsebiosciences.com/products/evercode-wt-mega/
95. Ch. 12 — Chan Zuckerberg Initiative. CZ CELLxGENE Discover Census: Census data releases. Web page (the long-term-support release of 8 November 2025). https://chanzuckerberg.github.io/cellxgene-census/cellxgene_census_docsite_data_release_info.html
96. Ch. 12 — Tahoe-100M: A Giga-Scale Single-Cell Perturbation Atlas for Context-Dependent Gene Function and Cellular Modeling. bioRxiv doi:10.1101/2025.02.20.639398, posted 24 February 2025 (preprint of Tahoe Therapeutics, formerly Vevo Therapeutics; its Crossref record gives 14 authors at Tahoe Therapeutics and 18 at Parse Biosciences)
97. Ch. 12 — 10x Genomics, Inc. Space Ranger glossary. Software documentation, undated (manufacturer: Visium spots 55 µm across; Visium HD squares 2 µm wide). https://www.10xgenomics.com/support/software/space-ranger/latest/getting-started/space-ranger-glossary
98. Ch. 12 — Philippidis A. Aiming high: 10x, NanoString launch large-scale panels. GEN, 2 July 2024 (trade press; the makers' panel figures). https://genengnews.com/topics/omics/aiming-high-10x-nanostring-launch-large-scale-panels
99. Ch. 12 — Bruker Spatial Biology. CosMx Human Whole Transcriptome Product Bulletin, MK6755. April 2025 (manufacturer: more than 18,000 RNAs). https://www.brukerspatialbiology.com/wp-content/uploads/2025/04/CosMx_Human_Whole_Transcriptome_Product_Bulletin.pdf
100. Ch. 13 — Sender R., Fuchs S., Milo R. Revised estimates for the number of human and bacteria cells in the body. PLoS Biology 14(8), e1002533 (2016). doi:10.1371/journal.pbio.1002533 (Weizmann Institute of Science)
101. Ch. 13 — Johnson J.S. et al. Evaluation of 16S rRNA gene sequencing for species and strain-level microbiome analysis. Nature Communications 10, 5029 (2019). doi:10.1038/s41467-019-13036-1
102. Ch. 13 — rrnDB (University of Michigan). About. Web page (copy numbers of the 16S gene by species). https://rrndb.umms.med.umich.edu/about
103. Ch. 13 — Benoit P. … Chiu C.Y. Seven-year performance of a clinical metagenomic next-generation sequencing test for diagnosis of central nervous system infections. Nature Medicine 30, 3522–3533 (2024). doi:10.1038/s41591-024-03275-1 (University of California San Francisco; five of the 20 authors gave Delve Bio as an affiliation, per the paper's OpenAlex record)
104. Ch. 13 — 360Dx. Seven-Year Study Validates Metagenomic Sequencing for Diagnosing CNS Infections. News report, November 2024. https://www.360dx.com/sequencing/seven-year-study-validates-metagenomic-sequencing-diagnosing-cns-infections (trade press; states that the senior author of Benoit et al. 2024 and colleagues co-founded Delve Bio in 2023, which licensed the technology from UCSF; five of the paper's 20 authors gave Delve Bio as an affiliation in 2024, per the paper's OpenAlex record)
105. Ch. 13 — Charalampous T. … O'Grady J. Nanopore metagenomics enables rapid clinical diagnosis of bacterial lower respiratory infection. Nature Biotechnology 37, 783–792 (2019). doi:10.1038/s41587-019-0156-5 (abstract read; University of East Anglia; part-funded by Oxford Nanopore Technologies, which also supplied flow cells; one author a full-time employee and share-option holder of the company)
106. Ch. 13 — Karius, Inc. Press release (Business Wire). 16 May 2024 (company statement: the test offered as a laboratory-developed test; Breakthrough Device designation for suspected lung infections in immunocompromised patients; title not recorded). https://www.businesswire.com/news/home/20240516554967/en
107. Ch. 14 — Schwarzenbach H. Circulating nucleic acids as biomarkers in breast cancer. Breast Cancer Research 15, 211 (2013). doi:10.1186/bcr3446 (review; half-life of circulating nucleic acids from 15 minutes to several hours)
108. Ch. 14 — American College of Obstetricians and Gynecologists, Committee on Practice Bulletins–Obstetrics, with the Society for Maternal-Fetal Medicine. Screening for Fetal Chromosomal Abnormalities. Practice Bulletin No. 226. Obstetrics & Gynecology 136(4), e48–e69 (October 2020). Read in a reproduction hosted by Natera (a test maker), https://www.natera.com/wp-content/uploads/2021/04/ACOG-PB-226.pdf
109. Ch. 14 — Sequenom Center for Molecular Medicine. Announces launch of MaterniT21 noninvasive prenatal test for Down syndrome. Press release (PR Newswire, republished by BioSpace). 17 October 2011 (company statement). https://www.biospace.com/b-sequenom-center-for-molecular-medicine-b-announces-launch-of-maternit21-tm-noninvasive-prenatal-test-for-down-syndrome
110. Ch. 14 — Diamandis E.P., Fiala C. Can circulating tumor DNA be used for direct and early stage cancer detection? F1000Research 6, 2129 (2017). https://f1000research.com/articles/6-2129 (opinion article with a model calculation; an estimate of the tumor share of plasma DNA, not a cohort measurement)
111. Ch. 14 — MDxHealth SA. Exo mdx (ExoDx Prostate) product page and Methods & Limitations page. Web pages, undated, read 24 October 2026 (manufacturer). https://mdxhealth.com/?p=34043 ; https://exo.prod.exo.mdxhealth.com/methods-limitations (a urine test of three genes in RNA from EVs)
112. Ch. 14 — NIST. Certificate of Analysis, Standard Reference Material 2372a, Human DNA Quantitation Standard. Issued 13 December 2024 (certified in human haploid genome equivalents per nanoliter and in nanograms of DNA per microliter, the mass calculated from the copy number; the ratio of the two gives about 3.3 pg per haploid genome equivalent). https://tsapps.nist.gov/srmext/certificates/2372a.pdf
113. Ch. 14 — Eurofins Genomics. Plasma-preparation for cell-free DNA extraction, v3.0 (2019) (a genomics service provider's customer instructions: about 4 to 5 mL of plasma from one 10 mL blood collection tube; a rule of thumb, not a measurement). https://www.eurofinsgenomics.eu/media/1610538/k-1300_plasma_preparation_en.pdf
114. Ch. 14 — Tsui W.H.A., Ding S.C., Jiang P., Lo Y.M.D. Artificial intelligence and machine learning in cell-free-DNA-based diagnostics. Genome Research 35(1) (2025). doi:10.1101/gr.278413.123 (review; modal cell-free DNA length about 166 bp, fetal and tumor-derived about 143 bp; Lo and Jiang declared equity in DRA, Take2 and Insighta, and Jiang in Illumina)
115. Ch. 14 — Hu X., Ding S.C., Jiang P. Emerging frontiers of cell-free DNA fragmentomics. Extracellular Vesicles and Circulating Nucleic Acids 3, 380–392 (2022). https://www.oaepublish.com/articles/evcna.2022.34 (review; Jiang declared equity in DRA and Grail/Illumina)
116. Ch. 14 — van der Pol E., Coumans F., Varga Z., Krumrey M., Nieuwland R. Innovation in detection of microparticles and exosomes. Journal of Thrombosis and Haemostasis 11 (Suppl 1), 36–45 (2013). doi:10.1111/jth.12254 (review; read in the copy on the first author's site; funded by the European Metrology Research Programme; the authors declared no conflict of interest). https://www.edwinvanderpol.com/downloads/articles/2013/van_der_Pol_2013_JTH_Innovation_in_detection.pdf
117. Ch. 14 — Černiauskas L., Aleknavičiūtė G., Arbačiauskaitė S., Mytzka N., Allelein S., Bikulčienė I., Karčiauskaitė D. High density lipoproteins and extracellular vesicles: distinct but overlapping circulating particles and their role in atherosclerosis. Frontiers in Cardiovascular Medicine 13 (2026), published 29 July 2026. doi:10.3389/fcvm.2026.1831078 (review; the authors stated that they received no funding and declared no conflict of interest). https://www.frontiersin.org/journals/cardiovascular-medicine/articles/10.3389/fcvm.2026.1831078/full
118. Ch. 14 — Society for Maternal-Fetal Medicine. Consult Series #74. 2025. doi:10.1002/pmf2.70139 (professional society guidance, endorsed by the American College of Obstetricians and Gynecologists; read on the society's Consult #74 page at publications.smfm.org and in its post of 1 December 2025 at smfm.org; title not recorded)
119. Ch. 14 — Malki Y., Kang G., Lam W.K.J., Zhou Q., … Jiang P., Chan K.C.A., Lo Y.M.D. Analysis of a cell-free DNA–based cancer screening cohort links fragmentomic profiles, nuclease levels, and plasma DNA concentrations. Genome Research 35(1), 31–42 (2025; online 27 November 2024). doi:10.1101/gr.279667.124 (abstract and back matter; as declared in 2024, Chan and Lo held equity in DRA, Take2 and Insighta and received patent income from companies including Grail and Illumina, Jiang held equity in Illumina, several authors were directors of DRA, Take2, KingMed Future and Insighta, and a patent application had been filed on the data)
120. Ch. 14 — US FDA. Shield (P230009), Guardant Health: Recently-Approved Devices page (approved 26 July 2024; the maker's study of 7,861 people), https://www.fda.gov/medical-devices/recently-approved-devices/shield-p230009 ; PMA database via openFDA (received 10 March 2023; supplement S010, a methylation-only classification model on the 180-day track, received 14 November 2025, approved 22 July 2026; data last updated 12 October 2026), https://api.fda.gov/device/pma.json?search=pma_number:P230009&sort=decision_date:asc&limit=3
121. Ch. 14 — Welsh J.A., Goberdhan D.C.I., O'Driscoll L., … Théry C., Witwer K.W.; MISEV Consortium. Minimal information for studies of extracellular vesicles (MISEV2023): From basic to advanced approaches. Journal of Extracellular Vesicles 13(2), e12404 (2024). doi:10.1002/jev2.12404 (the International Society for Extracellular Vesicles' position paper)
122. Ch. 14 — Welsh J.A., Van Der Pol E., Arkesteijn G.J.A. … Jones J.C. MIFlowCyt-EV: a framework for standardized reporting of extracellular vesicle flow cytometry experiments. Journal of Extracellular Vesicles 9(1), 1713526 (2020). doi:10.1080/20013078.2020.1713526 (abstract only)
123. Ch. 14 — Moss J. … Dor Y. Comprehensive human cell-type methylation atlas reveals origins of circulating cell-free DNA in health and disease. Nature Communications 9, 5068 (2018). doi:10.1038/s41467-018-07466-6
124. Ch. 14 — van der Pol E., Sturk A., van Leeuwen T., Nieuwland R., Coumans F.; ISTH-SSC-VB Working group. Standardization of extracellular vesicle measurements by flow cytometry through vesicle diameter approximation. Journal of Thrombosis and Haemostasis 16(6), 1236–1245 (2018). doi:10.1111/jth.14009 (van der Pol and Coumans shareholders in Exometry BV, which supplied the study's calibration system)
125. Ch. 14 — Bettegowda C., Sausen M., Leary R.J., … Papadopoulos N., Diaz L.A. Jr. Detection of circulating tumor DNA in early- and late-stage human malignancies. Science Translational Medicine 6(224), 224ra24 (2014). doi:10.1126/scitranslmed.3007094 (read in the author manuscript, the source of the 47 % at stage I, which the manuscript's Fig. 3B draws at about 44 %; the five senior authors owned stock in Personal Genome Diagnostics, two of them its co-founders, and Johns Hopkins held related patents, as declared in 2014; other authors at Life Technologies, Amgen, Indivumed and MyGenostics)
126. Ch. 14 — Kim J., Xu S., Jung S.-R., … Welsh J.A., Nolan J.P., Chiu D.T. Comparison of EV characterization by commercial high-sensitivity flow cytometers and a custom single-molecule flow cytometer. Journal of Extracellular Vesicles 13(8), e12498 (2024). doi:10.1002/jev2.12498 (Zhao M., Cheng Y., Fujimoto B.S. and Chiu D.T. declared a financial interest in Pangnostics, which licensed the single-molecule flow cytometer's technology from the University of Washington; one co-author at Micareo Inc.)
127. Ch. 14 — Bio-Techne Corporation. FDA Grants Breakthrough Device Designation To Bio-Techne's ExoDx Prostate IntelliScore (EPI) Test. Press release. 17 June 2019 (company statement, including the test's addition to a clinical guideline). https://investors.bio-techne.com/press-releases/detail/144/fda-grants-breakthrough-device-designation-to-bio-technes
128. Ch. 14 — EV-TRACK Consortium; Van Deun J. et al. EV-TRACK: transparent reporting and centralizing knowledge in extracellular vesicle research. Nature Methods (2017; online 28 February 2017). doi:10.1038/nmeth.4185 (abstract only; Ghent University; volume and pages not checked on the record)
129. Ch. 14 — NIST, Material Measurement Laboratory, Biosystems and Biomaterials Division. Biomolecular reference materials. Web page, accessed 24 October 2026 (no reference material for extracellular vesicles listed). https://www.nist.gov/mml/bbd/standards-and-tools/reference-materials
130. Ch. 15 — Li J.J., Bickel P.J., Biggin M.D. System wide analyses have underestimated protein abundances and the importance of transcription in mammals. PeerJ 2, e270 (2014). doi:10.7717/peerj.270
131. Ch. 15 — Vogel C., Marcotte E.M. Insights into the regulation of protein abundance from proteomic and transcriptomic analyses. Nature Reviews Genetics 13, 227–232 (2012). doi:10.1038/nrg3185 (abstract only)
132. Ch. 16 — US FDA. Quality Management System Regulation: Frequently Asked Questions. Web page, content current as of 2 October 2026. https://www.fda.gov/medical-devices/quality-management-system-regulation-qmsr/quality-management-system-regulation-frequently-asked-questions
133. Ch. 16 — Illumina, Inc. Phasing Correction (RTA2). NextSeq 500/550 support documentation, undated web page (manufacturer). https://support-docs.illumina.com/IN/NextSeq_550-500/Content/IN/PhasingCorrection_RTA2.htm
134. Ch. 16 — SCIEX. Continuing the data independent acquisition (r)evolution: Introducing ZT Scan DIA for quantitative proteomics. Technical note MKT-31731-A, 2024 (manufacturer). https://sciex.com/content/dam/SCIEX/pdf/tech-notes/technology/ztscan-dia-introduction.pdf
135. Ch. 16 — Jiang L. et al. Synthetic spike-in standards for RNA-seq experiments. Genome Research 21, 1543–1551 (2011). doi:10.1101/gr.121095.111 (abstract only)
136. Ch. 17 — Centers for Medicare & Medicaid Services. Clinical Laboratory Improvement Amendments (CLIA). Web page, last modified 7 October 2026. https://www.cms.gov/medicare/quality/clinical-laboratory-improvement-amendments
137. Ch. 17 — Illumina. Patterned Flow Cell Technology. Web page, undated (© 2026) (manufacturer). https://www.illumina.com/science/technology/next-generation-sequencing/sequencing-technology/patterned-flow-cells.html
138. Ch. 17 — DIN Media. ISO/DIS 25379-1:2026, In vitro diagnostic Next Generation Sequencing (NGS) workflows — Part 1: Human DNA examination, draft standard dated September 2026; and DIN EN ISO 25379-1:2026-11, draft issued 25 September 2026 (standards publisher's catalogue pages). https://www.dinmedia.de/en/draft-standard/iso-dis-25379-1/406650759 ; https://www.dinmedia.de/en/draft-standard/din-en-iso-25379-1/406368528
139. Ch. 17 — National Cancer Institute, Biorepositories and Biospecimen Research Branch. NCI Biospecimen Evidence-Based Practices: Cell-Free DNA: Biospecimen Collection and Processing. Revision 1.0. 5 February 2025 (first issued 30 January 2020). https://dctd.cancer.gov/data-tools-biospecimens/biospecimens-biobanks/resources/best-practices/bebp/bebp-cfdna
140. Ch. 17 — Centers for Medicare & Medicaid Services. 42 CFR 493.1253, Standard: Establishment and verification of performance specifications, in CMS's reproduction of Subpart K with interpretive guidelines. Undated. https://www.cms.gov/files/document/apcsubk1pdf
141. Ch. 17 — US FDA. Considerations for Design, Development, and Analytical Validation of Next Generation Sequencing (NGS)-Based In Vitro Diagnostics (IVDs) Intended to Aid in the Diagnosis of Suspected Germline Diseases. Final guidance. 13 April 2018. https://www.fda.gov/media/99208/download
142. Ch. 17 — ISO. ISO/DIS 25379-1, In vitro diagnostic Next Generation Sequencing (NGS) workflows — Part 1: Human DNA examination. Draft international standard, ballot opened 16 September 2026 (catalogue page). https://www.iso.org/standard/90120.html
143. Ch. 17 — ISO. ISO 20186-1, -2 and -3:2019, Molecular in vitro diagnostic examinations — Specifications for pre-examination processes for venous whole blood (Part 1 isolated cellular RNA; Part 2 isolated genomic DNA; Part 3 isolated circulating cell free DNA from plasma, confirmed 12 December 2024) (catalogue pages). https://www.iso.org/standard/69800.html
144. Ch. 17 — ISO. ISO 20184-1, -2 and -3 (frozen tissue: RNA, proteins, DNA; 2018 and 2021) and ISO 20166-1 to -4 (formalin-fixed and paraffin-embedded tissue; 2018 and 2021) (catalogue pages). https://www.iso.org/standard/67215.html
145. Ch. 17 — Illumina. BCL Convert Software Guide v3.7.5. Document 1000000163594 v00. April 2021 (manufacturer). https://support.illumina.com/content/dam/illumina-support/documents/documentation/software_documentation/bcl_convert/bcl-convert-v3-7-5-software-guide-1000000163594-00.pdf
146. Ch. 17 — Cock P.J.A., Fields C.J., Goto N., Heuer M.L., Rice P.M. The Sanger FASTQ file format for sequences with quality scores, and the Solexa/Illumina FASTQ variants. Nucleic Acids Research 38, 1767–1771 (2010). doi:10.1093/nar/gkp1137 (abstract only)
147. Ch. 17 — SAM/BAM Format Specification Working Group. Sequence Alignment/Map Format Specification. Version b5341fb, 12 August 2025. https://samtools.github.io/hts-specs/SAMv1.pdf
148. Ch. 17 — Global Alliance for Genomics and Health, Large Scale Genomics work stream. The Variant Call Format Specification: VCFv4.5 and BCFv2.2. 25 February 2026. https://samtools.github.io/hts-specs/VCFv4.5.pdf
149. Ch. 17 — HUPO Proteomics Standards Initiative. mzML (format page; version 1.1.0, 1 June 2009). https://www.psidev.info/mzML ; Martens L. et al. mzML—a community standard for mass spectrometry data. Molecular & Cellular Proteomics 10, R110.000133 (2011). doi:10.1074/mcp.R110.000133 (abstract only; authors include instrument-vendor staff)
150. Ch. 17 — Costello M. et al. Characterization and remediation of sample index swaps by non-redundant dual indexing on massively parallel sequencing platforms. BMC Genomics 19, 332 (2018). doi:10.1186/s12864-018-4703-0 (all authors at the Broad Institute; funded by NHGRI)
151. Ch. 17 — Pengelly R.J. et al. A SNP profiling panel for sample tracking in whole-exome sequencing studies. Genome Medicine 5, 89 (2013). doi:10.1186/gm492
152. Ch. 17 — NIST. Certificate of Analysis, Standard Reference Material 2374, DNA Sequence Library for External RNA Controls. Revised 6 December 2017. https://tsapps.nist.gov/srmext/certificates/2374.pdf
153. Ch. 17 — NIST. Certificate of Analysis, Standard Reference Material 1950, Metabolites in Frozen Human Plasma. Revised 12 October 2023. https://tsapps.nist.gov/srmext/certificates/1950.pdf
154. Ch. 17 — Geeurickx E. et al. The generation and use of recombinant extracellular vesicles as biological reference material. Nature Communications 10 (2019). doi:10.1038/s41467-019-11182-0 (abstract only; article number not verified)
155. Ch. 17 — Welsh J.A. et al. Towards defining reference materials for measuring extracellular vesicle refractive index, epitope abundance, size and concentration. Journal of Extracellular Vesicles 9, 1816641 (2020). doi:10.1080/20013078.2020.1816641 (abstract of the arXiv preprint 2004.09428 only)
156. Ch. 17 — Illumina. White paper on index hopping on patterned flow cells. Pub. No. 770-2017-004-D. 2018 (manufacturer; title not recorded). https://jp.illumina.com/content/dam/illumina-marketing/documents/products/whitepapers/index-hopping-white-paper-770-2017-004.pdf
157. Ch. 17 — Leek J.T. et al. Tackling the widespread and critical impact of batch effects in high-throughput data. Nature Reviews Genetics 11, 733–739 (2010). doi:10.1038/nrg2825 (abstract only; a review)
158. Ch. 17 — NIST. Report of Investigation, Reference Material 8398, Human DNA for Whole-Genome Variant Assessment (HG-001). Revised 13 December 2024. https://tsapps.nist.gov/srmext/certificates/8398.pdf
159. Ch. 18 — Benjamini Y., Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society, Series B 57(1), 289–300 (1995). doi:10.1111/j.2517-6161.1995.tb02031.x (abstract only)
160. Ch. 18 — Broadhurst D. et al. Guidelines and considerations for the use of system suitability and quality control samples in mass spectrometry assays applied in untargeted clinical metabolomic studies. Metabolomics 14, 72 (2018). doi:10.1007/s11306-018-1367-3
161. Ch. 19 — Tartal J. (US FDA, CDRH). Design Controls. Training slides, undated (on the superseded Quality System Regulation, 21 CFR 820.30). https://www.fda.gov/media/116762/download
162. Ch. 19 — IEC. IEC 60812:2018, Failure modes and effects analysis (FMEA and FMECA). Edition 3.0, 10 August 2018 (catalogue page). https://webstore.iec.ch/en/publication/26359
163. Ch. 19 — IEC. IEC 61025:2006, Fault tree analysis (FTA). Edition 2.0, 13 December 2006 (catalogue page). https://webstore.iec.ch/publication/4311
164. Ch. 19 — Wilbon T.A. (US FDA, CDRH). Quality Management System Regulation (QMSR) – Risk Management, Risk-Based Approach, and Risk-Based Decisions. Training presentation, undated. https://www.fda.gov/media/189040/download
165. Ch. 19 — Johner Institute. SOUP (definition of software of unknown provenance, reproducing IEC 62304 clause 3.29). Web page, undated (a company that sells regulatory training and consulting; the standard itself not read). https://johner-institute.com/index.php?id=133
166. Ch. 19 — IEC SC 62A/MT 49. 62304 2nd Edition Change Rationales, cover letter. Committee document N0166, undated (SOUP subclauses 5.3.3 and 5.3.4 of the current edition). https://assets.iec.ch/public/sc62a/N0166_IEC62304_ED2_ChangeRationales_CoverLetter.pdf
167. Ch. 19 — US FDA. Evaluation of Automatic Class III Designation for Invitae Common Hereditary Cancers Panel. De Novo decision summary DEN210011, granted 29 September 2023 (precision study, Section VI.A.1 and Table 5; the maker's study as FDA reports it). https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN210011.pdf ; De Novo database record https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?id=DEN210011
168. Ch. 19 — SEQC/MAQC-III Consortium. A comprehensive assessment of RNA-seq accuracy, reproducibility and information content by the Sequencing Quality Control Consortium. Nature Biotechnology 32, 903–914 (2014). doi:10.1038/nbt.2957 (abstract and acknowledgements only; coordinated by FDA; authors include staff of Illumina, Life Technologies and other gene-expression product companies; volume and pages not checked on the record)
169. Ch. 19 — Office of the Federal Register. 21 CFR Part 820, Quality Management System Regulation (§§ 820.1–820.45), eCFR, up to date as of 21 October 2026; final rule 89 FR 7496, 2 February 2024. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-820
170. Ch. 19 — ISO. ISO 13485:2016, Medical devices — Quality management systems — Requirements for regulatory purposes. Edition 3, confirmed 31 October 2025 (catalogue page). https://www.iso.org/standard/59752.html
171. Ch. 19 — Tartal J. (US FDA, CDRH). Quality Management System Regulation (QMSR) – Design and Development. Training presentation, undated. https://www.fda.gov/media/189041/download
172. Ch. 19 — ISO. ISO 14971:2019, Medical devices — Application of risk management to medical devices. Edition 3, confirmed 7 March 2025 (catalogue page). https://www.iso.org/standard/72704.html
173. Ch. 19 — ISO. ISO/TR 24971:2020, Medical devices — Guidance on the application of ISO 14971 (catalogue page); ISO/AWI TS 24971-1, approved work item registered 6 July 2026 (catalogue page). https://www.iso.org/standard/74437.html ; https://www.iso.org/standard/94297.html
174. Ch. 19 — International Medical Device Regulators Forum, SaMD Working Group. Characterization Considerations for Medical Device Software and Software-Specific Risk. IMDRF/SaMD WG/N81 FINAL:2025. 27 January 2025. https://www.imdrf.org/sites/default/files/2025-01/IMDRF_SaMD%20WG_Software-Specific%20Risk_N81%20Final_0.pdf
175. Ch. 19 — Medical Device Coordination Group. MDCG 2025-10, Guidance on post-market surveillance of medical devices and in vitro diagnostic medical devices. December 2025 (guidance, not legally binding). https://health.ec.europa.eu/document/download/a9ad86b7-1b8e-4bae-beb4-48b2b3ed2f05_en?filename=mdcg_2025-10_en.pdf
176. Ch. 19 — US FDA. Applying Human Factors and Usability Engineering to Medical Devices. Guidance for Industry and FDA Staff, issued 3 August 2026 (first issued 3 February 2016). https://www.fda.gov/media/80481/download
177. Ch. 19 — IEC. IEC 62304:2006+AMD1:2015, Medical device software — Software life cycle processes. Edition 1.1 (catalogue page and preview). https://webstore.iec.ch/en/publication/6792
178. Ch. 19 — IEC SC 62A. Design Specification for the second edition of IEC 62304 (proposal, undated). https://assets.iec.ch/public/sc62a/IEC%2062304%20Ed.%202.0%20Design%20Specification.pdf
179. Ch. 19 — US FDA. Content of Premarket Submissions for Device Software Functions. Final guidance. 14 June 2023. https://www.fda.gov/media/153781/download
180. Ch. 19 — Centers for Medicare & Medicaid Services. LDT and CLIA FAQs. Version 2013.10.22. https://www.cms.gov/regulations-and-guidance/legislation/clia/downloads/ldt-and-clia_faqs.pdf
181. Ch. 19 — CLSI. EP05 (4th edition, 11 December 2025), EP06 (2nd edition, 24 November 2020), EP07 (3rd edition, 30 April 2018, reaffirmed October 2022), EP09 (3rd edition, 20 June 2018), EP12 (3rd edition, 7 March 2023), EP17-A2 (2nd edition, 18 June 2012, reaffirmed June 2017) and MM09 (3rd edition, 20 April 2023) (catalogue pages only). https://clsi.org/shop/standards/
182. Ch. 19 — Aziz N. et al. College of American Pathologists' laboratory standards for next-generation sequencing clinical tests. Archives of Pathology & Laboratory Medicine 139, 481–493 (2015). doi:10.5858/arpa.2014-0250-CP (abstract only)
183. Ch. 19 — ISO. ISO 15189:2022, Medical laboratories — Requirements for quality and competence. Edition 4, December 2022 (catalogue page). https://www.iso.org/standard/76677.html
184. Ch. 19 — Gupta R., Agarwal R. Translating global standards to local practice: comparing ISO 15189:2012 and 2022 with implementation insights from NABL 112 in India. Journal of Laboratory Physicians 18, 175–183 (2026). doi:10.25259/JLP_316_2025 (secondary review)
185. Ch. 19 — National Accreditation Board for Testing and Calibration Laboratories (NABL). News and announcements: transition to ISO 15189:2022 (29 December 2022); NABL 112A and 112B (18 December 2024); implementation by 31 December 2025. https://nabl-india.org/?p=737
186. Ch. 19 — Health Canada. GUI-0079, glossary reproducing the ISO 14971:2019 definitions of hazard, hazardous situation, harm and risk. 21 November 2025 (a regulator's reproduction; the standard itself not read).
187. Ch. 19 — Tartal J. (US FDA, CDRH). Risk Basics For Medical Devices. Training slides, undated (reproducing ISO 14971:2019 Annex C, Figure C.1: hazard, sequence of events, hazardous situation, harm). https://fda.gov/media/163912/download
188. Ch. 20 — AAMI. ANSI/AAMI/ISO 14971:2019, Medical devices — Application of risk management to medical devices. Preview pages (cover, front matter and clauses 1 to 3.1; the contents give clause 10 as "Production and post-production activities"; the clause text not read). https://webstore.ansi.org/preview-pages/AAMI/preview_ANSI+AAMI+ISO+14971-2019.pdf
189. Ch. 20 — Rosenzweig C. Third edition of ISO 14971 – What is changing. Johner Institute blog, 25 August 2022, modified 10 February 2025 (a company that sells regulatory training and consulting; its summary of clause 10, subclauses 10.1 to 10.4; the standard itself not read). https://blog.johner-institute.com/iso-14971-risk-management/third-edition-of-iso-14971/
190. Ch. 20 — Koul S. Medical device bodies welcome CDSCO order on self-reporting adverse effects. Business Standard, New Delhi, 21 May 2024 (news report: a CDSCO circular dated 15 May urged licensed makers to report adverse events through the MvPI portal; the circular itself not read). https://www.business-standard.com/health/medical-device-bodies-welcome-cdsco-order-on-self-reporting-adverse-effects-124052101276_1.html
191. Ch. 20 — JSA Advocates & Solicitors. JSA Newsletter, Healthcare, May & June 2024 Edition (law firm; secondary: a CDSCO circular of 15 May 2024 to license holders on adverse events, who are "suggested to use the MvPI platform"; the circular itself not read). https://www.jsalaw.com/newsletters-and-updates/jsa-newsletter-healthcare-may-june-2024-edition/
192. Ch. 20 — Boldrick A., Pugh E. (Emergo by UL). Vigilance Reporting in the European Economic Area, Switzerland and the United Kingdom. White paper, June 2024 (regulatory consultancy; secondary: counts EU vigilance limits in calendar days with the awareness date as day 0, and names no source for the rule). https://www.emergobyul.com/sites/default/files/2024-08/RLC24CS1445442-Vigilance_Reporting-Whitepaper.pdf
193. Ch. 20 — Council of the European Communities. Regulation (EEC, Euratom) No 1182/71 of 3 June 1971 determining the rules applicable to periods, dates and time limits, Articles 1 and 3, read in the revised UK version published by legislation.gov.uk (Article 3(1), (3) and (4): the day of the event left out, public holidays, Sundays and Saturdays included, a period ending on one of them extended to the next working day; whether the regulation governs IVDR vigilance deadlines was not found in any source read). https://legislation.gov.uk/eur/1971/1182/article/3?view=plain ; https://www.legislation.gov.uk/eur/1971/1182
194. Ch. 20 — US FDA (openFDA). Device recall record and enforcement record Z-1927-2026 (event 98625): Foundation Medicine, Sequencing Agent SEQ0067, a component of FoundationOne CDx; lots 24346, 24386 and 24488, 163 units; reason "Possible increase in phasing, potentially resulting in false-positive BARD1 splice site indel artifact variants"; voluntary, firm initiated on 20 March 2026; classified Class II on 22 April 2026; root cause "Under Investigation by firm"; open. Data last updated 22 October 2026 (openFDA states its data are unvalidated). https://api.fda.gov/device/recall.json?search=product_res_number:"Z-1927-2026" ; https://api.fda.gov/device/enforcement.json?search=recall_number:"Z-1927-2026"
195. Ch. 20 — US FDA, Center for Devices and Radiological Health. PMA Supplements and Amendments. Web page, content current as of 10 December 2025 (describes the 180-day supplement and the 30-day notice of 21 CFR 814.39). https://www.fda.gov/medical-devices/premarket-approval-pma/pma-supplements-and-amendments
196. Ch. 20 — US FDA (CDRH, ORA/OII). Compliance Program 7382.850, Inspection of Medical Device Manufacturers. Issued 2 February 2026, implementation date 2 February 2026 (supersedes CP 7382.845 and CP 7383.001; QMS areas "Outsourcing and Purchasing", "Measurement, Analysis, and Improvement" and "Change Control"; read in part, Attachment A not read). https://www.fda.gov/media/80195/download
197. Ch. 20 — Wilbon T. (US FDA, CDRH). Navigating the Quality Management System Regulation. Training presentation, undated (comparison table of the Quality System Regulation and the QMSR: purchasing controls and corrective and preventive action incorporated by reference). https://www.fda.gov/media/181437/download
198. Ch. 20 — US FDA. Medical Devices; Quality System Regulation Amendments. Final rule, Federal Register 89(23), 7496 ff., 2 February 2024, as posted by FDA (preamble read in part: Comments 12, 20 and 33; in force since 2 February 2026). https://www.fda.gov/media/177022/download
199. Ch. 20 — The National Archives (legislation.gov.uk). Regulation (EU) 2017/746, Annex IX (conformity assessment based on a quality management system and on assessment of technical documentation; sections 2.4, 3.4 and 4.11 read): the original version as adopted by the EU, reproduced by a government publisher; not checked against the consolidated text. https://www.legislation.gov.uk/eur/2017/746/annex/IX/adopted
200. Ch. 20 — Ministry of Health and Family Welfare. Medical Devices Rules, 2017, G.S.R. 78(E), 31 January 2017, in force 1 January 2018; CDSCO's copy as amended to G.S.R. 777(E) of 14 October 2022 (rules 3, 20, 21, 26, 31, 32, 34, 36, 38, 40, 41, 51, 52, 59, 63 and 64 read; the Fifth and Sixth Schedules not reached; amendments after 14 October 2022 not checked). https://cdsco.gov.in/opencms/resources/UploadCDSCOWeb/2022/m_device/mdr,%202017%20(1).pdf
201. Ch. 20 — CDSCO. Addendum to Doc No. CDSCO/IVD/FAQ/04/2022, Addendum No. 02. 13 March 2026. https://www.cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/Addendum-Doc-No-CDSCOIVDFAQ-042022dated-13032026.pdf
202. Ch. 20 — US FDA, Center for Devices and Radiological Health. Mandatory Reporting Requirements: Manufacturers, Importers and Device User Facilities. Web page, content current as of 27 March 2025 (FDA's summary of 21 CFR Part 803; the regulation itself was not read on the eCFR). https://www.fda.gov/medical-devices/postmarket-requirements-devices/mandatory-reporting-requirements-manufacturers-importers-and-device-user-facilities
203. Ch. 20 — European Parliament and Council. Regulation (EU) 2017/746 on in vitro diagnostic medical devices (IVDR). 5 April 2017, OJ L 117, 5 May 2017 (original text; Articles 2 and 5(5) read, among them the definitions of incident, serious incident, corrective action, field safety corrective action and field safety notice, Article 2(67), (68) and (70) to (72); Annex VIII, Rule 3, not re-read, the text extraction ending before the Annexes). https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32017R0746
204. Ch. 20 — The National Archives (legislation.gov.uk). Regulation (EU) 2017/746, Article 82 (content: the maker's reports of serious incidents and field safety corrective actions, and their time limits; article title not recorded): the original version as adopted by the EU, reproduced by a government publisher; not checked against the consolidated text. https://www.legislation.gov.uk/eur/2017/746/article/82/adopted
205. Ch. 20 — The National Archives (legislation.gov.uk). Regulation (EU) 2017/746, Article 84 (content: the maker's investigation of serious incidents, the field safety notice and the competent authority's evaluation; article title not recorded): the original version as adopted by the EU, reproduced by a government publisher; not checked against the consolidated text. https://www.legislation.gov.uk/eur/2017/746/article/84/adopted
206. Ch. 20 — Lexplosion Solutions. Compliance news item on CDSCO's circular on reporting suspected unexpected serious adverse events to the Materiovigilance Programme of India (MvPI). 3 June 2024 (secondary: a compliance-software firm's summary, title not recorded; the CDSCO circular it reports, dated 15 May 2024 by its file name, could not be read). https://lexplosion.in/medical-device-license-holders-mandated-to-report-all-suspected-unexpected-serious-adverse-events-to-the-materiovigilance-programme-of-india-mvpi-for-timely-risk-identification-and-regulatory-compli/
207. Ch. 20 — Health Products Regulatory Authority (Ireland). Mandatory use of EUDAMED modules for medical devices. Notice. 23 March 2026 (national competent authority). https://www.hpra.ie/news-events/news/article/mandatory-use-of-eudamed-modules-for-medical-devices
208. Ch. 20 — US FDA. PMA database, P250029: SimpleScreen CRC, Freenome, LLC; received 1 August 2025, approved 24 July 2026; supplement S001, a 30-day notice of "changes in a supplier manufacturing processes for critical reagents" (reason "Process Change - Manufacturer/Sterilizer/Packager/Supplier"), received 27 July 2026, decided 26 August 2026 (decision code OK30: the notice accepted, which is not an approval). https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P250029 ; openFDA https://api.fda.gov/device/pma.json?search=pma_number:P250029&limit=5
209. Ch. 20 — US FDA. Medical Device Reporting for Manufacturers. Guidance for Industry and Food and Drug Administration Staff, 8 November 2016 (final, after a draft of 9 July 2013; supersedes the guidance of March 1997; no withdrawal seen as of October 2026; sections 2.1, 2.3, 2.5, 2.13, 2.14, 2.20, 2.21 and 4.4.1 read). https://www.fda.gov/media/86420/download
210. Ch. 20 — Office of the Federal Register. 21 CFR Part 806, Medical devices; reports of corrections and removals (§§ 806.1, 806.2, 806.10 and 806.20 read). eCFR, up to date as of 21 October 2026. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-806
211. Ch. 20 — Medical Device Coordination Group. MDCG 2022-2, Guidance on general principles of clinical evidence for In Vitro Diagnostic medical devices (IVDs). January 2022 (guidance, not legally binding; performance evaluation and its post-market performance follow-up). https://health.ec.europa.eu/system/files/2022-01/mdcg_2022-2_en.pdf
212. Ch. 21 — Sasieni P., Johnson P., Round T., Warwick J., Jones H., Dolly S., Lee L., Lee Y., Kumar H., Liang W., Smittenaar R., Neal R.D., Swanton C. Effect of Screening with Multicancer Early-Detection Test on Late-Stage Cancer Diagnosis. New England Journal of Medicine, published online 22 September 2026. doi:10.1056/NEJMoa2505723 (read through the NEJM Research Summary and the abstract; funded by GRAIL; Kumar, Liang and Smittenaar affiliated with GRAIL Bio UK; volume and pages not yet assigned)
213. Ch. 21 — US FDA. PMA database, P130001: Epi proColon, methylated Septin 9 DNA in plasma, approved 12 April 2016 (approval order addressed to Epigenomics AG; the record now names New Day Diagnostics, LLC as applicant). https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P130001
214. Ch. 21 — Centers for Medicare & Medicaid Services. National Coverage Determination 90.2, Next Generation Sequencing (NGS). Medicare Coverage Database; version 1 effective 16 March 2018, version 2 effective 27 January 2020. https://www.cms.gov/medicare-coverage-database/view/ncd.aspx?ncdid=372
215. Ch. 21 — Palmetto GBA. MolDX: Coverage, Coding, and Pricing Standards and Requirements (M00106), version 26.0. December 2019 (manual of a Medicare Administrative Contractor; whether it is the current version was not checked). https://palmettogba.com/Palmetto/Moldx.Nsf/files/MolDX_Manual.pdf/$File/MolDX_Manual.pdf
216. Ch. 21 — Centers for Medicare & Medicaid Services. Clinical Laboratory Fee Schedule (CLFS). Web page, last modified 24 September 2026. https://www.cms.gov/medicare/payment/fee-schedules/clinical-laboratory-fee-schedule-clfs
217. Ch. 21 — Regulation (EU) 2021/2282 of the European Parliament and of the Council of 15 December 2021 on health technology assessment. Official Journal of the European Union L 458, 22 December 2021, p. 1 (Articles 1(2) and 7). https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32021R2282
218. Ch. 21 — Unnithan R. (Member of Parliament, Kasaragod). Regarding Cancer care in India. Statement in Parliament, Parliament Digital Library, undated in the extract (a member's statement, not a government answer). https://eparlib.sansad.in/bitstream/123456789/2991308/1/3046.pdf
219. Ch. 21 — MTRC. Companion Diagnostic Tests. Web page, undated (a market-access consultancy: "Due to their high cost and novelty, CDx tests are often the subject of HTA"). https://mtrconsult.com/node/2816
220. Ch. 21 — Institute of Medicine. Evolution of Translational Omics: Lessons Learned and the Path Forward. National Academies Press (2012), consensus study report 13297 (requested by the National Cancer Institute after the Duke trials; the project page read, the report itself not opened). https://www.nationalacademies.org/projects/IOM-HCS-10-06/publication/13297
221. Ch. 21 — US FDA. PMA database, P170019: FoundationOne CDx, Foundation Medicine, approved 30 November 2017; P190032: FoundationOne Liquid CDx, approved 26 August 2020; P200010: Guardant360 CDx, approved 7 August 2020, with supplements S024 (ESR1, camizestrant; 4 September 2026) and S029 (ESR1, imlunestrant with abemaciclib; 18 September 2026); P230011: TruSight Oncology Comprehensive, Illumina, approved 21 August 2024. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P170019 ; https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P190032 ; https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P200010 ; https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P230011 ; openFDA, data last updated 12 October 2026, https://api.fda.gov/device/pma.json?search=applicant:Guardant&sort=decision_date:desc&limit=10
222. Ch. 21 — US FDA. 510(k) decision summary K081092, MammaPrint (Agendia), a modification of the 70-gene signature, citing its earlier decisions and the classification regulation 21 CFR 866.6040. https://www.accessdata.fda.gov/cdrh_docs/reviews/K081092.pdf
223. Ch. 21 — Wikipedia. MammaPrint. Encyclopedia article, read 24 October 2026 (secondary: the De Novo route of February 2007 and the description of the signature as the first multigene test of its kind, citing Simon, Paik and Hayes, Journal of the National Cancer Institute, 2009, not opened).
224. Ch. 21 — Sparano J.A. et al. (TAILORx; ECOG-ACRIN). New England Journal of Medicine 379, 111–121 (2018). doi:10.1056/NEJMoa1804710 (bibliographic record only; funded by the National Cancer Institute)
225. Ch. 21 — National Cancer Institute. TAILORx trial results press release. 3 June 2018 (the trial's funder). https://www.cancer.gov/news-events/press-releases/2018/tailorx-breast-cancer-chemotherapy
226. Ch. 21 — Cardoso F. et al. 70-gene signature as an aid to treatment decisions in early-stage breast cancer. New England Journal of Medicine 375, 717–729 (2016). doi:10.1056/NEJMoa1602253 (abstract only; two authors at Agendia, the signature's maker, per the Europe PMC record)
227. Ch. 21 — Hegi M.E. et al. MGMT gene silencing and benefit from temozolomide in glioblastoma. New England Journal of Medicine 352, 997–1003 (2005). doi:10.1056/NEJMoa043331 (not opened)
228. Ch. 21 — US FDA. Reclassification of Nucleic Acid-Based Test Systems for Use with a Corresponding Approved Oncology Therapeutic Product. Proposed order, Federal Register document 2025-21071, docket FDA-2025-N-4622 (public inspection copy; proposed 21 CFR 866.6075; product codes OWD, PJG, PQP, SFL; cites the 2014 guidance In Vitro Companion Diagnostic Devices). 25 November 2025. https://public-inspection.federalregister.gov/2025-21071.pdf
229. Ch. 21 — Office of the Federal Register. 21 CFR 864.1890, in situ hybridization test systems for use with a corresponding approved oncology therapeutic product (final order, 91 FR 53190, 17 August 2026; product codes MVD, NYQ, OWE, PNK). eCFR, up to date as of 8 October 2026. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-864/subpart-B/section-864.1890
230. Ch. 21 — Discoveries in Health Policy (policy blog; author and title not recorded). Post on the FDA's reclassification of companion diagnostics. 3 September 2026 (secondary: the ISH order's effective date of 16 September 2026 and its proposal of 11 June 2025). https://www.discoveriesinhealthpolicy.com/2026/09/fda-is-really-doing-it-downclassify-cdx.html
231. Ch. 21 — US FDA. PMA database, P260004: Signatera CDx, Natera; companion diagnostic for atezolizumab in muscle-invasive bladder cancer; product code PQP; received 29 January 2026, approved 15 May 2026; the only PMA record under Natera's name in openFDA (data last updated 12 October 2026). https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpma/pma.cfm?id=P260004 ; https://api.fda.gov/device/pma.json?search=applicant:Natera&sort=decision_date:desc&limit=5
232. Ch. 21 — US FDA. PMA database, P250027: Guardant360 Liquid CDx, Guardant Health; 741 genes; product code PQP; received 31 July 2025, approved 19 May 2026. openFDA. https://api.fda.gov/device/pma.json?search=pma_number:P250027&limit=10
233. Ch. 21 — Clark M.M. et al. Meta-analysis of the diagnostic and clinical utility of genome and exome sequencing and chromosomal microarray in children with suspected genetic diseases. npj Genomic Medicine 3, 16 (2018). doi:10.1038/s41525-018-0053-8 (abstract only; the senior author, Kingsmore S.F., then at Rady Children's Institute for Genomic Medicine, which offers rapid genome sequencing)
234. Ch. 21 — Chung C.C.Y. et al. Meta-analysis of the diagnostic and clinical utility of exome and genome sequencing in pediatric and adult patients with rare diseases across diverse populations. Genetics in Medicine 25(9), 100896 (2023). doi:10.1016/j.gim.2023.100896 (abstract only)
235. Ch. 21 — 100,000 Genomes Project Pilot Investigators. 100,000 Genomes pilot on rare-disease diagnosis in health care, preliminary report. New England Journal of Medicine 385(20), 1868–1880 (2021) (abstract only, read in the Institute of Cancer Research repository record). https://repository.icr.ac.uk/handle/internal/5462
236. Ch. 21 — Kingsmore S.F., Nofsinger R., Ellsworth K. Rapid genomic sequencing for genetic disease diagnosis and therapy in intensive care units: a review. npj Genomic Medicine 9 (2024). doi:10.1038/s41525-024-00404-0 (abstract only; the authors at Rady Children's Institute for Genomic Medicine, which offers rapid genome sequencing)
237. Ch. 21 — Genomics England. First newborn babies tested for over 200 genetic conditions as world leading study begins in NHS hospitals. News. 3 October 2024 (the program's own announcement: up to 100,000 newborns; conditions treatable in early childhood). https://www.genomicsengland.co.uk/news/first-newborn-babies-tested-for-over-200-genetic-conditions-as-world-leading-study-begins-in-nhs-hospitals
238. Ch. 21 — Genomics England. 25,000 babies join groundbreaking Generation Study. News. 23 October 2025 (the program's own announcement). https://www.genomicsengland.co.uk/news/25-000-babies-join-groundbreaking-generation-study
239. Ch. 21 — NewYork-Presbyterian. News release on the first results of the GUARDIAN newborn genome-sequencing study. 5 November 2024 (the study hospital's release; title and URL not recorded: results from the study's first 4,000 newborns, published in JAMA on 24 October 2024, the paper not opened; "Funding for the study was provided by Sanofi, Illumina, GeneDx, and the Columbia Precision Medicine Initiative"). nyp.org
240. Ch. 21 — Clinical Pharmacogenetics Implementation Consortium. CPIC guidelines. Web page. https://cpicpgx.org/guidelines/ (redirects to https://www.clinpgx.org/cpic/guidelines; the page renders by script, and its list was not read)
241. Ch. 21 — US FDA. Table of Pharmacogenomic Biomarkers in Drug Labeling. Web page, last modified 12 August 2026. https://www.fda.gov/drugs/science-and-research-drugs/table-pharmacogenomic-biomarkers-drug-labeling
242. Ch. 21 — Carss K., Halldorsson B.V. et al. (UK Biobank Whole-Genome Sequencing Consortium). Whole-genome sequencing of 490,640 UK Biobank participants. Nature 645 (2025), published 6 August 2025. doi:10.1038/s41586-025-09272-9 (abstract only; the consortium includes company scientists)
243. Ch. 21 — All of Us Research Program (NIH). Curated Data Repository (CDR) version 9 Release Notes. Web page. https://support.researchallofus.org/hc/en-us/articles/50653909888788-Curated-Data-Repository-CDR-version-9-Release-Notes
244. Ch. 22 — Queen Mary University of London. Major NHS trial provides new evidence on multi-cancer blood test. News. 23 September 2026 (the investigators' university: who ran the trial; GRAIL funded it). https://www.qmul.ac.uk/news/latest-news/2026/medicine-and-dentistry/fmd/major-nhs-trial-provides-new-evidence-on-multi-cancer-blood-test.html
245. Ch. 22 — Neal R.D. et al. Performance of a multi-cancer early detection test in the randomized controlled NHS-Galleri trial. Nature Medicine 32(10), 3770–3780 (2026). doi:10.1038/s41591-026-04652-8 (the same 13 authors as the NEJM paper, three of them affiliated with GRAIL; funding statement not read)
246. Ch. 22 — Petricoin E.F. et al. Use of proteomic patterns in serum to identify ovarian cancer. Lancet 359, 572–577 (2002). PMID 11867112; doi:10.1016/S0140-6736(02)07746-2 (abstract read through the Early Detection Research Network's publication record; affiliations from the OpenAlex record: six of the 11 authors at the US National Cancer Institute and FDA, and two at Correlogic Systems, the company that planned to offer the test)
247. Ch. 22 — Sullivan M.G. Controversy erupts over proteomics studies. MDedge ObGyn, 15 April 2005 (trade press). https://www.mdedge.com/obgyn/article/51470/gynecology/controversy-erupts-over-proteomics-studies/page/0/1
248. Ch. 22 — Office of Research Integrity, HHS. Findings of Research Misconduct. Federal Register 2015-28437, 9 November 2015 (abstract read through the Federal Register listing). https://www.federalregister.gov/documents/2015/11/09/2015-28437/findings-of-research-misconduct
249. Ch. 22 — Centers for Disease Control and Prevention. ACCE model for evaluating genetic tests. Web page. https://www.cdc.gov/genomics/gtesting/acce/index.htm (the page could not be opened)
250. Ch. 22 — LabMedica. News report of 1 June 2026 on GRAIL's presentation of the full NHS-Galleri results at ASCO on 30 May 2026 (news report of a company presentation; the source of GRAIL's "nominally significant"). https://mobile.labmedica.com/latest-news/articles/294811011/grail-presents-full-results-from-nhs-galleri-trial-at-asco-2026.html
251. Ch. 22 — US FDA, Center for Devices and Radiological Health. September 23, 2026: Molecular and Clinical Genetics Panel of the Medical Devices Advisory Committee Meeting Announcement. Web page, content current as of 24 September 2026 (docket FDA-2026-N-8004; a public meeting of the panel to discuss, make recommendations on and vote on GRAIL's PMA). https://www.fda.gov/advisory-committees/advisory-committee-calendar/september-23-2026-molecular-and-clinical-genetics-panel-medical-devices-advisory-committee-meeting
252. Ch. 22 — US FDA. Marketing Submission Recommendations for a Predetermined Change Control Plan for AI-Enabled Device Software Functions. Final guidance, notice of availability, Federal Register 2024-28361, 4 December 2024 (abstract read through the Federal Register listing). https://www.federalregister.gov/documents/2024/12/04/2024-28361/marketing-submission-recommendations-for-a-predetermined-change-control-plan-for-artificial
253. Ch. 22 — CDSCO, Medical Devices & IVD Division. Guidance Document on Medical Device Software, Doc No. CDSCO/MD/GD/MDSW/01/2026. Undated document, 62 pages (issued by circular on 21 July 2026 per three secondary reports). https://www.cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/Guidance-document-on-Medical-Device-Software-under-MDR-2017.pdf
254. Ch. 22 — Nishith Desai Associates. From Code to Compliance: Decoding India's Medical Device Software Guidance. 3 August 2026 (legal commentary; secondary, one of the three reports that give the guidance's circular date of 21 July 2026). https://nishithdesai.com/research-and-articles/hotline/pharma-healthcare-update/from-code-to-compliance-decoding-indias-medical-device-software-guidance-15623
255. Ch. 22 — Pure Global. India CDSCO Medical Device Software Guidance 2026. Undated (consultancy; secondary). https://www.pureglobal.com/news/india-cdsco-medical-device-software-guidance-2026
256. Ch. 22 — Emergo by UL. India CDSCO finalizes guidance on medical device software. 30 July 2026 (consultancy; secondary; dates the draft to 2025). https://www.emergobyul.com/news/india-cdsco-finalizes-guidance-medical-device-software
257. Ch. 23 — Jones Day. MGI Tech Sells Complete Genomics to Swiss Rockets. Cases & Deals. February 2026 (the seller's counsel: Complete Genomics as MGI Tech's US subsidiary). https://www.jonesday.com/en/practices/experience/2026/02/mgi-tech-sells-complete-genomics-to-swiss-rockets
258. Ch. 23 — Covington & Burling LLP. BIOSECURE Act Moves Closer to Enactment with Inclusion in FY 2026 NDAA Text. Inside Government Contracts (blog). 10 December 2025 (legal commentary on the text released 7 December 2025, before enactment). https://www.insidegovernmentcontracts.com/2025/12/biosecure-act-moves-closer-to-enactment-with-inclusion-in-fy-2026-ndaa-text/
259. Ch. 23 — US Department of Defense, Office of the Under Secretary of Defense (Acquisition and Sustainment). Notice of Availability of Designation of Chinese Military Companies. Federal Register 91, 35189–35194, 10 June 2026 (the section 1260H list, dated 5 June 2026: BGI Group with Complete Genomics among its subsidiaries; MGI Tech). https://www.govinfo.gov/content/pkg/FR-2026-06-10/html/2026-11571.htm
260. Ch. 23 — Ultima Genomics, Inc. Ultima Genomics Launches Paired-End Sequencing and Solaris 2.0 Max at ASHG 2026, Doubling Output and Enabling the $60 Genome. Press release (PR Newswire), Montréal. 20 October 2026 (company statement: up to 60,000 30x genomes a year on one UG200 Ultra; what the $60 covers is not stated). https://www.prnewswire.com/news-releases/ultima-genomics-launches-paired-end-sequencing-and-solaris-2-0-max-at-ashg-2026--doubling-output-and-enabling-the-60-genome-302914208.html
261. Ch. 23 — Seer, Inc. Proteograph Technology. Web page, undated (manufacturer: engineered nanoparticles, protein corona, compressed dynamic range). https://seer.bio/technology/ ; Wiegel S. High-Throughput Proteomics Workflow. Genetic Engineering & Biotechnology News, 1 July 2025 (trade press relaying the maker: "nanoparticle-enabled enrichment"). https://www.genengnews.com/?p=318554
262. Ch. 23 — US FDA. Establishment registration and device listing via openFDA, proprietary name FACSLyric (owner/operator Becton Dickinson and Company for every site listed). Data last updated 12 October 2026. https://api.fda.gov/device/registrationlisting.json?search=proprietary_name:FACSLyric&limit=5
263. Ch. 23 — 10x Genomics, Inc. 10x Genomics Begins Global Shipments of Atera, Putting Unprecedented Spatial Biology Capabilities in Researchers' Hands. Press release (PR Newswire, reproduced by Finviz), Pleasanton, Calif. 29 September 2026 (company statement; no price or research-use statement). https://finviz.com/news/396563/10x-genomics-begins-global-shipments-of-atera-putting-unprecedented-spatial-biology-capabilities-in-researchers-hands
264. Ch. 23 — Illumina, Inc. Illumina completes acquisition of SomaLogic. Press release, San Diego. 30 January 2026 (company statement; $350 million in cash at closing and up to $75 million in performance-based milestones and royalties; Illumina Protein Prep renamed SomaSeq Discovery). https://www.illumina.com/company/news-center/press-releases/2026/b7bf54f5-b534-47b4-a629-02ade3b10178.html
265. Ch. 23 — Thermo Fisher Scientific Inc. Thermo Fisher Scientific Completes Acquisition of Olink, Announces Commencement of Subsequent Offering Period. Press release (Business Wire, via BioSpace). 10 July 2024 (company statement). https://www.biospace.com/thermo-fisher-scientific-completes-acquisition-of-olink-announces-commencement-of-subsequent-offering-period
266. Ch. 23 — Bruker Corporation. Bruker Completes Asset Acquisition of NanoString Business. Press release (Business Wire, via BioSpace). 6 May 2024 (company statement; no separate closing date given). https://www.biospace.com/bruker-completes-asset-acquisition-of-nanostring-business
267. Ch. 23 — Quanterix Corporation. Form 10-Q for the quarter ended 30 June 2026 (SEC filing; acquisition of Akoya Biosciences completed 8 July 2025; PhenoCycler and PhenoImager listed as Quanterix trademarks). https://www.sec.gov/Archives/edgar/data/0001503274/000150327426000036/qtrx-20260630.htm
268. Ch. 23 — MDxHealth SA. Mdxhealth Announces Closing of Exosome Diagnostics Acquisition from Bio-Techne. Press release (GlobeNewswire). 15 September 2025 (company statement); Bio-Techne Corporation. Bio-Techne Announces Exosome Diagnostics Divestiture. Press release. 5 August 2025 (company statement; Bio-Techne retains access to the exosome-based technology for kit development). https://mdxhealth.com/?p=33592 ; https://investors.bio-techne.com/news/detail/498/bio-techne-announces-exosome-diagnostics-divestiture
269. Ch. 23 — QIAGEN N.V. Form 6-K, quarterly report for the period ended 30 June 2026, Note 3 (SEC filing; acquisition of Parse Biosciences completed 2 December 2025; announcement date not stated). https://www.sec.gov/Archives/edgar/data/0001015820/000101582026000053/a2026q2quarterlyreport.htm
270. Ch. 23 — Waters Corporation. Waters Completes Combination with BD's Biosciences & Diagnostic Solutions Businesses. Press release (PR Newswire, via BioSpace). 9 February 2026 (company statement; names no instrument). https://www.biospace.com/press-releases/waters-completes-combination-with-bds-biosciences-diagnostic-solutions-businesses
271. Ch. 23 — Abbott Laboratories. Abbott completes acquisition of Exact Sciences. Press release (PR Newswire, via Nasdaq), Abbott Park. 23 March 2026 (company statement; gives no regulatory status for Oncodetect or Cancerguard). https://www.nasdaq.com/press-release/abbott-completes-acquisition-exact-sciences-2026-03-23
272. Ch. 23 — RTTNews. CareDx Closes Acquisition Of Naveris For Approx. $260 Mln. 2 July 2026 (news report; the company's release not opened; no regulatory status given for NavDx). https://www.rttnews.com/3664226/caredx-closes-acquisition-of-naveris-for-approx-260-mln-gains-navdx-mrd-test-for-cancer.aspx
273. Ch. 23 — 10x Genomics, Inc. 10x Genomics Acquires Proteintech Genomics, a Division Within Proteintech Group. Press release. 9 June 2026 (company statement; terms not disclosed; closing not stated). https://investors.10xgenomics.com/news/news-details/2026/10x-Genomics-Acquires-Proteintech-Genomics-a-Division-Within-Proteintech-Group/default.aspx
274. Ch. 23 — Complete Genomics, Inc. Complete Genomics Enters Definitive Agreement to be Acquired by Swiss Rockets AG. News release. 23 February 2026 (company statement); Complete Genomics Appoints Lauren Selig to Board of Directors. News release, 19 August 2026, via 01net.it (company statement: approved by MGI Tech shareholders, still subject to CFIUS clearance). https://www.completegenomics.com/complete-genomics-enters-definitive-agreement-to-be-acquired-by-swiss-rockets-ag/ ; https://www.01net.it/complete-genomics-appoints-lauren-selig-to-board-of-directors/
275. Ch. 23 — Standard BioTools Inc. Form 8-K, Exhibit 99.1. 30 September 2026 (SEC filing; company statement: agreement to sell the Mass Cytometry business to Element Biosystems, closing expected by the end of 2026); Form 8-K, Exhibit 99.1, 28 July 2026 (the earlier agreement with Multiplex Bio, since ended; the pending merger with Treeline Biosciences). https://www.sec.gov/Archives/edgar/data/0001162194/000114036126038110/ef20082995_ex99-1.htm ; https://www.sec.gov/Archives/edgar/data/0001162194/000114036126029841/ef20078901_ex99-1.htm
276. Ch. 23 — Office of the Federal Register. 21 CFR 809.10, Labeling for in vitro diagnostic products. eCFR, up to date as of 20 October 2026 (authoritative but unofficial). https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-809/subpart-B/section-809.10
277. Ch. 23 — US FDA. 510(k) database via openFDA, applicant Illumina: De Novo DEN130011 (MiSeqDx Platform) and DEN130042 (MiSeqDx Universal Kit 1.0), and 510(k)s K124006 and K132750 (MiSeqDx cystic fibrosis assays), all decided 19 November 2013. Data last updated 12 October 2026. https://api.fda.gov/device/510k.json?search=applicant:Illumina&sort=decision_date:desc&limit=15
278. Ch. 23 — Element Biosciences, Inc. Element Biosciences Introduces VITARI, Redefining What High-Throughput Sequencing Makes Possible. Press release. 19 February 2026 (company statement; the "first" and the $100 genome are the company's; AVITI24 spatial sequencing appears on the company's site as a plan). https://www.elementbiosciences.com/news/element-biosciences-introduces-vitari-redefining-what-high-throughput-sequencing-makes-possible
279. Ch. 23 — Ultima Genomics, Inc. Ultima Genomics Introduces the Next Generation of Sequencing at Scale with the Launch of the UG200 Series and Next-Generation Solaris 2.0 Workflows. Press release reproduced as sponsored content by The Scientist. February 2026 (company statement). https://www.the-scientist.com/ultima-genomics-introduces-the-next-generation-of-sequencing-at-scale-with-the-launch-of-the-ug200-series-and-next-generation-solaris-2-0-workflows-74133
280. Ch. 23 — Complete Genomics, Inc. News listing, page 2: DNBSEQ-T1+ and DNBSEQ-E25 Flash unveiled at AGBT (23 February 2025); commercial partnership with the Human Cell Atlas for Stereo-seq spatial transcriptomics (25 March 2025) (company headlines). https://www.completegenomics.com/category/news/page/2/
281. Ch. 23 — Pacific Biosciences of California, Inc. Press release on the NMPA Class III registration of the Sequel II CNDx system with Berry Genomics. 4 November 2025 (company statement, including the "world's first"; title not recorded; the NMPA record not opened). https://pacb.com/press_releases/pacbio-supports-berry-genomics-in-achieving-first-regulatory-approval-for-clinical-long-read-sequencing-in-china/
282. Ch. 23 — Oxford Nanopore Technologies plc. Oxford Nanopore Receives Regulatory Approval for its First Diagnostic Device in the UK and Europe. Press release (Business Wire). January 2026 (company statement; date from the release identifier; device class and notified body not stated). https://www.businesswire.com/news/home/20260111466198/en/oxford-nanopore-receives-regulatory-approval-for-its-first-diagnostic-device-in-the-uk-and-europe
283. Ch. 23 — Illumina, Inc. Infinium MethylationEPIC v2.0 Kit. Product page, undated (manufacturer). https://www.illumina.com/products/by-type/microarray-kits/infinium-methylation-epic.html
284. Ch. 23 — New England Biolabs, Inc. NEBNext Enzymatic Methyl-seq v2 Kit (E8015). Product page, undated (manufacturer). https://www.neb.com/en-us/products/e8015-nebnext-enzymatic-methyl-seq-v2-kit
285. Ch. 23 — Illumina, Inc. Press release on the 5-base solution for variants and methylation from one sample. 15 October 2025 (company statement; title not recorded). https://www.illumina.com/company/news-center/press-releases/2025/ed906867-8ddf-4219-b15f-b8aa8facbd11.html
286. Ch. 23 — GenomeWeb. ASMS Product Launches: Thermo Fisher Scientific, Bruker, SCIEX, Seer, Evosep. 2025 (trade press reporting the makers' figures; year from the meeting's venue, Baltimore). https://www.genomeweb.com/new-products/asms-product-launches-thermo-fisher-scientific-bruker-sciex-seer-evosep/639371
287. Ch. 23 — Lab Manager. Innovative New Products Introduced at ASMS 2026. Undated (trade press reporting the makers' figures); SelectScience. Top mass spectrometry launches at ASMS 2026. 11 June 2026 (trade press). https://www.labmanager.com/innovative-new-products-introduced-at-asms-2026-35617 ; https://www.selectscience.net/article/top-mass-spectrometry-launches-at-asms-2026
288. Ch. 23 — Alamar Biosciences, Inc. News page and product navigation: NULISAseq panels; profiling of about 21,000 plasma samples with the Neuro 220 panel for the CLARiTI consortium, announced 13 July 2026 (company statements; products marked for research use only). https://www.alamarbio.com/news/
289. Ch. 23 — Nautilus Biotechnology, Inc. Nautilus Biotechnology Reports Second Quarter 2026 Financial Results. Press release, 28 July 2026, via Santelog (company statement). https://www.santelog.com/actualites-sante-nasdaq/nautilus-biotechnology-reports-second-quarter-2026-financial-results
290. Ch. 23 — Metabolon, Inc. Home page (accessed 24 October 2026; company statements, including the "world's largest" reference library; research use only). https://www.metabolon.com/
291. Ch. 23 — biocrates life sciences gmbh. Home page (modified 3 September 2026; company statements; research use only). https://www.biocrates.com/
292. Ch. 23 — US FDA. 510(k) database via openFDA: K170974, BD FACSLyric Flow Cytometer system with FACSuite Clinical software and BD Multitest assays (3 July 2017), and K201814, with the BD FACSDuet Sample Preparation System and further optical configurations (28 September 2020); applicant Becton, Dickinson and Company. Data last updated 12 October 2026. https://api.fda.gov/device/510k.json?search=device_name:FACSLyric+OR+device_name:CytoFLEX+OR+device_name:FACSDuet&sort=decision_date:desc&limit=10
293. Ch. 23 — Bruker Spatial Biology, Inc. Home page (accessed 24 October 2026; manufacturer; "For Research Use Only. Not for use in diagnostic procedures."). https://www.brukerspatialbiology.com/
294. Ch. 23 — 10x Genomics, Inc. 10x Genomics Introduces Atera, a New Platform to Redefine How Biology is Measured and Understood. Press release. 18 April 2026 (company statement; no price or research-use statement). https://investors.10xgenomics.com/news/news-details/2026/10x-Genomics-Introduces-Atera-a-New-Platform-to-Redefine-How-Biology-is-Measured-and-Understood/default.aspx
295. Ch. 23 — Vizgen, Inc. News page (accessed 24 October 2026; company headlines, including a report of $48 million raised, 8 January 2026). https://vizgen.com/news/
296. Ch. 23 — Freenome, Inc. Form 8-K. 23 July 2026 (SEC filing; business combination with Perceptive Capital Solutions Corp completed 20 July 2026; exhibits include a collaboration and license agreement with Exact Sciences Corporation dated 3 August 2025, its terms not described). https://www.sec.gov/Archives/edgar/data/0002017526/000114036126029471/ef20078324_8k.htm
297. Ch. 23 — US FDA. PMA database via openFDA, applicant Exact Sciences: P130017, Cologuard (received 7 June 2013, approved 11 August 2014), and P230043, Cologuard Plus (received 22 December 2023, approved 3 October 2024). Data last updated 12 October 2026. https://api.fda.gov/device/pma.json?search=applicant:Exact+AND+supplement_number:%22%22&sort=decision_date:desc&limit=10
298. Ch. 23 — US FDA. 510(k) Premarket Notification database, K260235: CellDx-Tissue, Datar Cancer Genetics Private Limited; Traditional 510(k), 21 CFR 866.6080, product code PZM; received 26 January 2026, decided 12 May 2026. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K260235
299. Ch. 23 — Novo Holdings A/S. MedGenome raises Series E funding co-led by Maj Invest and Novo Holdings to expand access to its genomics offerings in India. News release. 21 July 2025 (investor's release carrying MedGenome's own figures and its "largest" claim). https://novoholdings.com/news/medgenome-raises-series-e-funding-co-led-by-maj-invest-and-novo-holdings-to-expand-access-to-its-genomics-offerings-in-india
300. Ch. 23 — Strand Life Sciences. Website (accessed 24 October 2026; company statements). https://us.strandls.com/
301. Ch. 23 — 4baseCare. Home page (accessed 24 October 2026; company statements). https://www.4basecare.com/
302. Ch. 23 — Mapmygenome. Home page (accessed 24 October 2026; company statements, including NABL accreditation and ISO 15189:2012 certification). https://mapmygenome.in/
303. Ch. 23 — Datar Cancer Genetics. Home page (accessed 24 October 2026; company statements, including "India's only US FDA-cleared comprehensive genomic profiling assay provider" and its laboratories in Nashik and Guildford, UKAS Medical Laboratory No. 22238). https://datarpgx.com/
304. Ch. 23 — Public Law 119-60, National Defense Authorization Act for Fiscal Year 2026, approved 18 December 2025; division A, title VIII, section 851, "Prohibition on contracting with certain biotechnology providers" (govinfo; the section's number and title as the law's table of contents gives them). https://www.govinfo.gov/content/pkg/PLAW-119publ60/html/PLAW-119publ60.htm
305. Ch. 23 — Ropes & Gray LLP. BIOSECURE Act Enacted. Client alert. January 2026 (legal commentary on section 851 of the National Defense Authorization Act for Fiscal Year 2026, Public Law 119-60). https://www.ropesgray.com/en/insights/alerts/2026/01/biosecure-act-enacted
306. Ch. 23 — Element Biosciences, Inc. VITARI pre-order page ("Shipping in H2 2026") and news listing to 19 October 2026, read 24 October 2026 (company statements; no shipment announced, no price or research-use statement on the page). https://www.elementbiosciences.com/products/preorder-vitari ; https://www.elementbiosciences.com/news
307. Ch. 24 — Schwarze K., Buchanan J., Fermont J.M., … Taylor J.C., Wordsworth S. The complete costs of genome sequencing: a microcosting study in cancer and rare diseases from a single center in the United Kingdom. Genetics in Medicine 22(1), 85–94 (2020; published online 30 July 2019). doi:10.1038/s41436-019-0618-7 (sequencing consumables 68 to 72 % of the total cost; genetic counseling not costed; two authors received travel support from Illumina, and one honoraria and in-kind contributions from Illumina, Oxford Nanopore and others)
308. Ch. 24 — Technology Assessment at SickKids (TASK), The Hospital for Sick Children, Toronto. Microcosting of Whole Genome Sequencing (WGS) of Trios in a Heterogeneous Pediatric Cardiac Population. Report No. 2020-01, Highlights, 21 September 2020 (supplies 50.8 % and bioinformatics 24.8 % of the costs; clinical consultations and genetic counseling not captured). https://lab.research.sickkids.ca/task/wp-content/uploads/sites/66/2020/09/2020-01-CGC-HIGHLIGHTS.pdf
309. Ch. 24 — Christensen K.D., Dukhovny D., Siebert U., Green R.C. Assessing the Costs and Cost-Effectiveness of Genomic Sequencing. Journal of Personalized Medicine 5(4), 470–486 (2015). doi:10.3390/jpm5040470 (review; expects variant interpretation costs to fall more slowly than the cost of generating sequence; one author's research supported by Illumina grants, with compensation from Invitae, Illumina, Helix, Roche and others)
310. Ch. 24 — Avsec Ž., Latysheva N., Cheng J. et al. Advancing regulatory variant effect prediction with AlphaGenome. Nature 649, 1206–1218 (2026), published 28 January 2026 (Google DeepMind authors; read in the Nature press release of 29 January 2026, https://www.natureasia.com/en/info/press-releases/detail/9222)
311. Ch. 24 — Google DeepMind. AlphaGenome: AI for better understanding the genome. Blog, 25 June 2025. https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/ (company statement)
312. Ch. 24 — Brixi G., Durrant M.G., Ku J., Poli M. et al. Genome modeling and design across all domains of life with Evo 2. Manuscript, Arc Institute. https://arcinstitute.org/manuscripts/Evo2.pdf (authors at the Arc Institute, Stanford University, NVIDIA and others)
313. Ch. 24 — Abramson J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024), published 8 May 2024. doi:10.1038/s41586-024-07487-w (bibliographic record only; 48 authors at Google DeepMind and Isomorphic Labs)
314. Ch. 24 — Ahlmann-Eltze C., Huber W., Anders S. Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines. Nature Methods 22, 1657–1661 (2025). doi:10.1038/s41592-025-02772-6 (abstract only; EMBL and Heidelberg University)
315. Ch. 24 — Arc Institute. Virtual Cell Challenge 2025. News, 26 June 2025 (300,000 cells and 300 gene knockouts; industry sponsors). https://arcinstitute.org/news/virtual-cell-challenge-2025 ; GenomeWeb. Inaugural Virtual Cell Challenge reveals shortcomings of current AI cellular prediction. News report, December 2025. https://genomeweb.com/informatics/inaugural-virtual-cell-challenge-reveals-shortcomings-current-ai-cellular-prediction
316. Ch. 24 — Shift Bioscience. Shift Bioscience publication increases confidence in AI virtual cells for novel target discovery. Company announcement (SelectScience), 6 October 2026 (company statement, on its study with collaborators: Miller H.E., Mejia G.M., Leblanc F.J.A. et al. Deep learning perturbation models can outperform baselines on calibrated metrics. Nature Biotechnology (2026), not opened; URL not recorded)
317. Ch. 24 — GEN. New calibration framework improves confidence in AI virtual cell models. 6 October 2026 (news report, trade press, on the Shift Bioscience paper; URL not recorded)
318. Ch. 25 — US FDA, Center for Devices and Radiological Health. 24 Hour Summary: Molecular and Clinical Genetics Panel of the Medical Devices Advisory Committee Meeting. Meeting summary. 23 September 2026 (votes on Galleri: safety 10 to 0, effectiveness 6 to 4, benefit and risk 7 to 2 with one abstention; a summary, not the final record). https://www.fda.gov/media/195032/download
319. Ch. 25 — US FDA, Center for Devices and Radiological Health. General Controls for Medical Devices. Web page, content current as of 30 January 2026. https://www.fda.gov/medical-devices/regulatory-controls/general-controls-medical-devices
320. Ch. 25 — US FDA, Center for Devices and Radiological Health. Breakthrough Devices Program. Web page, modified 31 August 2026 (eligibility criteria; 1,320 designations and 210 marketing authorizations as of 30 June 2026). https://www.fda.gov/medical-devices/how-study-and-market-your-device/breakthrough-devices-program
321. Ch. 25 — Office of the Federal Register. 42 CFR 493.17 (the seven criteria for categorizing a test by complexity; a test not on the lists is of high complexity until reviewed). eCFR, up to date as of 15 October 2026. https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-G/part-493/subpart-A/section-493.17
322. Ch. 25 — Office of the Federal Register. 42 CFR 493.25 (the certificate and requirements for laboratories performing tests of high complexity). eCFR, up to date as of 27 August 2026. https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-G/part-493/subpart-A/section-493.25
323. Ch. 25 — Office of the Federal Register. 42 CFR Part 493, Subpart M (personnel for tests of moderate and high complexity, sections 493.1403 to 493.1495). eCFR, up to date as of 19 October 2026. https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-G/part-493/subpart-M
324. Ch. 25 — US FDA (openFDA). Original PMA approvals whose generic name contains "colorectal": three blood tests, approved in 2016, 2024 and 2026, and three stool tests. Data last updated 12 October 2026 (openFDA states its data are unvalidated). https://api.fda.gov/device/pma.json?search=generic_name:colorectal+AND+supplement_number:""&limit=20
325. Ch. 25 — US FDA. FDA Advisory Committee Meeting Reminder: Molecular and Clinical Genetics Panel – September 23, 2026. GovDelivery bulletin. 21 September 2026. https://content.govdelivery.com/accounts/USFDA/bulletins/42b39a2
326. Ch. 25 — US FDA. Medical Devices; Laboratory Developed Tests; Implementation of Vacatur. Final rule, 90 FR 45134, RIN 0910-AJ05, docket FDA-2025-N-1730. 19 September 2025 (implements the vacatur in American Clinical Laboratory Association v. FDA, No. 4:24-CV-479-SDJ, E.D. Tex., 31 March 2025). https://www.govinfo.gov/content/pkg/FR-2025-09-19/pdf/2025-18239.pdf
327. Ch. 25 — US FDA. Laboratory Developed Tests. Web page, content current as of 19 September 2025 (read 24 October 2026). https://www.fda.gov/medical-devices/vitro-diagnostics/laboratory-developed-tests
328. Ch. 25 — US FDA. Final rule adding laboratories as manufacturers of in vitro diagnostic products to 21 CFR 809.3(a), 89 FR 37286. 6 May 2024 (vacated 31 March 2025; cited from the 2025 rule and FDA's page, not opened; title not recorded).
329. Ch. 25 — US FDA. Overview of IVD Regulation. Web page, content current as of 20 December 2024 (quotes 21 CFR 809.3; describes the 510(k), De Novo and PMA routes). https://www.fda.gov/medical-devices/ivd-regulatory-assistance/overview-ivd-regulation
330. Ch. 25 — Office of the Federal Register. 21 CFR 860.3, Definitions (device classes). eCFR, up to date as of 20 October 2026. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-H/part-860/subpart-A/section-860.3
331. Ch. 25 — US FDA. Precision Medicine (page listing the two final guidances of April 2018: Considerations for Design, Development, and Analytical Validation of Next Generation Sequencing (NGS)-Based In Vitro Diagnostics (IVDs) Intended to Aid in the Diagnosis of Suspected Germline Diseases; Use of Public Human Genetic Variant Databases to Support Clinical Validity for Genetic and Genomic-Based In Vitro Diagnostics). Web page, content current as of 27 September 2018. https://www.fda.gov/medical-devices/in-vitro-diagnostics/precision-medicine
332. Ch. 25 — US FDA. Notices of availability of the two final NGS guidances, Federal Register 2018-07687 and 2018-07686, 13 April 2018 (abstracts read through the Federal Register listing; voluntary recognition of public variant databases). https://www.federalregister.gov/documents/2018/04/13/2018-07687/considerations-for-design-development-and-analytical-validation-of-next-generation-sequencing-based
333. Ch. 25 — OncNursingNews. FDA advisory panel votes to approve GRAIL Galleri MCED screening test. 27 September 2026 (secondary: Breakthrough Device designation in 2018; the PMA submitted 29 January 2026). https://www.oncnursingnews.com/view/fda-advisory-panel-votes-to-approve-grail-galleri-mced-screening-test
334. Ch. 25 — Office of the Federal Register. 42 CFR 493.5, Categories of tests by complexity. eCFR, up to date as of 29 September 2026. https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-G/part-493/subpart-A/section-493.5
335. Ch. 25 — US FDA, Center for Devices and Radiological Health. CLIA Categorizations. Web page, content current as of 17 July 2023. https://www.fda.gov/medical-devices/ivd-regulatory-assistance/clia-categorizations
336. Ch. 25 — US FDA. Quality Management System Regulation (QMSR). Web page, content current as of 2 February 2026 (QSIT retired on 2 February 2026; inspections under Compliance Program 7382.850). https://www.fda.gov/medical-devices/postmarket-requirements-devices/quality-management-system-regulation-qmsr
337. Ch. 25 — 360Dx. Grail Preparing for FDA Advisory Committee Meeting as Q2 Revenue Increases 26 Percent. August 2026 (secondary, reporting the company's expectation of approval in the "first part of 2027" and its data package). https://www.360dx.com/node/114696
338. Ch. 25 — GRAIL, Inc. FDA Advisory Committee Votes In Favor of Approval of GRAIL's Galleri Multi-Cancer Early Detection Test. Press release. 23 September 2026 (company statement; the same vote counts as the FDA's summary; the company's statement that it submitted the PMA on 29 January 2026 and that the FDA is expected to decide in the coming months). https://grail.com/press-releases/fda-advisory-committee-votes-in-favor-of-approval-of-grails-galleri-multi-cancer-early-detection-test/
339. Ch. 26 — Medical Device Coordination Group. MDCG 2020-16 rev.4, Guidance on Classification Rules for in vitro Diagnostic Medical Devices under Regulation (EU) 2017/746. March 2025 (EU guidance, not legally binding; reproduces Annex VIII and states that Rules 1 and 2 take precedence over Rule 3). https://health.ec.europa.eu/system/files/2023-02/md_mdcg_2020_guidance_classification_ivd-md_en.pdf
340. Ch. 26 — European Parliament and Council. Regulation (EU) 2017/746, consolidated version of 10 January 2025 (Articles 17 and 18, on the EU declaration of conformity and the CE marking, read; a consolidated text is a documentation tool, the Official Journal text is authentic). https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:02017R0746-20250110
341. Ch. 26 — The National Archives (legislation.gov.uk). Regulation (EU) 2017/746, Article 48 (conformity assessment by class): the original version as adopted by the EU, reproduced by a government publisher. https://www.legislation.gov.uk/eur/2017/746/article/48/adopted
342. Ch. 26 — European Parliament and Council. Regulation (EU) 2022/112 of 25 January 2022, amending the IVDR's transitional provisions and deferring conditions of its Article 5(5) (not opened; its content as the two summaries that follow give it).
343. Ch. 26 — Van Bael & Bellis. Regulation on In Vitro Diagnostic Medical Devices: transitional provisions and deferred application. 31 January 2022 (law firm; secondary for Regulation (EU) 2022/112). https://vbb.com/insights/corporate-commercial-regulatory/regulation-on-in-vitro-diagnostic-medical-devices-transitional-provisions-and-deferred-application
344. Ch. 26 — Eikenberg O., Loh E. (Emergo by UL). EU update: deadline nears for in-house devices to comply with IVDR. 1 March 2024 (regulatory consultancy; secondary; written before Regulation (EU) 2024/1860). https://www.emergobyul.com/news/eu-update-deadline-nears-house-devices-comply-ivdr
345. Ch. 26 — European Parliament and Council. Regulation (EU) 2024/1860 of 13 June 2024 amending Regulations (EU) 2017/745 and (EU) 2017/746 as regards a gradual roll-out of Eudamed, the obligation to inform in case of interruption or discontinuation of supply, and the transitional provisions for certain in vitro diagnostic medical devices. OJ L, 9 July 2024. https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=OJ%3AL_202401860
346. Ch. 26 — Baker McKenzie, Healthcare & Life Sciences blog. The EU's 2025 proposal to simplify the medical and in vitro diagnostic devices regulations (MDR/IVDR). 19 December 2025 (legal commentary; secondary for the proposal's content, which was not read). https://healthcarelifesciences.bakermckenzie.com/2025/12/19/the-eus-2025-proposal-to-simplify-the-medical-and-in-vitro-diagnostic-devices-regulations-mdr-ivdr/
347. Ch. 26 — European Commission. Proposal for a Regulation amending Regulations (EU) 2017/745, (EU) 2017/746, (EU) 2022/123 and (EU) 2024/1689 as regards simplifying and reducing the burden of the rules on medical devices and in vitro diagnostic medical devices, COM(2025) 1023 final, procedure 2025/0404(COD). 16 December 2025 (proposal; not adopted as of October 2026). EUR-Lex procedure page: https://eur-lex.europa.eu/procedure/EN/2025_404 ; Austrian Parliament EU database: https://www.parlament.gv.at/gegenstand/XXVIII/EU/64372
348. Ch. 26 — Council of the European Union. Regulation to simplify rules on medical and in vitro diagnostic devices – Progress report, 9801/26. 5 June 2026. https://data.consilium.europa.eu/doc/document/ST-9801-2026-INIT/en/pdf
349. Ch. 26 — Council of the European Union. Notice of meeting and provisional agenda, Working Party on Pharmaceuticals and Medical Devices, 26–27 October 2026, CM 4464/26 (third Presidency compromise, doc. 13649/26). 12 October 2026. https://data.consilium.europa.eu/doc/document/CM-4464-2026-INIT/en/pdf
350. Ch. 26 — Fieldfisher. Insight on the Commission's simplification proposal (title not recorded; legal commentary; secondary: adoption "anticipated by Q2 2027"). Undated. https://www.fieldfisher.com/en/locations/belgium/insights/breaking-the-bottleneck-the-commission-s-proposal
351. Ch. 26 — Council of the European Union. Cover note 9402/26 transmitting the opinion of the European Economic and Social Committee (adopted 29 April 2026: 191 for, 0 against, 4 abstentions). 19 May 2026. https://data.consilium.europa.eu/doc/document/ST-9402-2026-INIT/en/pdf
352. Ch. 27 — CDSCO. List of In-Vitro Diagnostic Medical Devices (updated), under rule 4(2) of the Medical Devices Rules, 2017, File No. IVD/Misc/196/2020. 25 October 2023 per CDSCO's IVD page (no date in the document). https://www.cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/medical-device/IVD_classifiction25oc23.pdf
353. Ch. 27 — CDSCO. Frequently Asked Questions on In-Vitro Diagnostic Medical Devices, CDSCO/IVD/FAQ/04/2022 (the current edition on CDSCO's IVD page, with an addendum of 28 March 2025; Q17 on products for research use only). https://www.cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/FAaddendum.pdf
354. Ch. 27 — Indian Council of Medical Research. Handbook on National Ethical Guidelines for Biomedical and Health Research Involving Human Participants. 2018 (ICMR's own summary of the 2017 guidelines; Section 10 on p. 14). https://ethics.ncdirindia.org/asset/pdf/Handbook_on_ICMR_Ethical_Guidelines.pdf
355. Ch. 27 — Parliament of India. The Pre-Conception and Pre-Natal Diagnostic Techniques (Prohibition of Sex Selection) Act, 1994 (Act No. 57 of 1994, as amended by Act 14 of 2003; sections 2, 3, 4, 5, 6 and 18 read in the copy published by the PC-PNDT division of the Ministry of Health and Family Welfare, the administering ministry; India Code not reachable). https://pndt.mohfw.gov.in/WriteReadData/l892s/PC-PNDT%20ACT-1994.pdf
356. Ch. 27 — CDSCO. In Vitro Diagnostics. Web page listing the IVD classification lists (23 July 2021; updated 25 October 2023), the provision for risk classification on the online system (6 February 2026), the ICMR-CDSCO Standard Performance Evaluation Protocols (17 September 2025), the compendium and IVD kit validation portal (18 September 2025), the draft guidance for import of IVDs (30 January 2026), the addendum to the IVD FAQ (13 March 2026), the updated list of performance-evaluation laboratories (29 June 2026) and the Draft Guidance document for Manufacturers of In-vitro diagnostic Medical device for stakeholders comment (9 October 2026); titles and dates read, documents not opened. https://www.cdsco.gov.in/opencms/opencms/en/Medical-Device-Diagnostics/InVitro-Diagnostics/
357. Ch. 27 — S.S. Rana & Co. Class C And D Non-notified Medical Devices To Require Licensing Under Medical Devices Rules 2017 From October 1, 2023. 10 October 2023 (legal commentary; secondary for the notification of 11 February 2020, G.S.R. 102(E), and the CDSCO notification of 12 April 2023, which were not opened). https://ssrana.in/articles/class-c-and-d-non-notified-medical-devices-to-require-licensing-under-medical-devices-rules-2017-from-october-1-2023/
358. Ch. 27 — CDSCO. Frequently Asked Questions on In Vitro Diagnostic Medical Devices, CDSCO/IVD/FAQ/03/2022 (FAQ 4, 17 and 96). 2022 (undated document). https://cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/IVD/FAQs/CDSCO-IVD-FAQ-03-2022-.pdf
359. Ch. 27 — WHO Prequalification Unit. In Vitro Diagnostics Eligible for WHO Prequalification. Web page, tables last updated 27 September 2026. https://extranet.who.int/prequal/vitro-diagnostics/vitro-diagnostics-eligible-who-prequalification
360. Ch. 27 — ILAC. ISO 15189:2022 for medical labs published. 8 December 2022. https://ilac.org/latest_ilac_news/iso-151892022-for-medical-labs-published/
361. Ch. 27 — Indian Council of Medical Research. National Ethical Guidelines for Biomedical and Health Research Involving Human Participants. October 2017 (contents and scope read; the body of Section 10 not read). https://ethics.ncdirindia.org/asset/pdf/ICMR_National_Ethical_Guidelines.pdf
362. Ch. 27 — CDSCO. Latest Public Notices ("Guidance document on Medical Device Software under MDR-2017" listed under 21 July 2026). Web page, accessed 26 October 2026. https://cdsco.gov.in/opencms/opencms/en/Latest-Public-Notices/