Noncoding Genome Function

Speculative If a large share of the 98.5% of the human genome that does not code for protein turns out to carry sequence-specific information — a claim that three independent lines of measurement currently put at between 3% and 11% of the genome rather than at 80%, and section 7 says what would settle it — then the number of interpretable positions in a clinical genome rises from about 30 million to several hundred million. The condition comes first because the answer is not known and the field’s two camps have both published their arguments in full.

Established The magnitude is best stated as the size of the uninterpretable fraction, because that is the quantity a diagnostic laboratory actually meets. Protein-coding exons are about 1.5% of 3.1 billion base pairs. The Zoonomia alignment of 240 placental mammals estimates that at least 10.7% of the human genome is evolutionarily constrained — roughly 332 megabases — and that 80% of the 101 million bases it can call constrained at single-base resolution lie outside protein-coding exons. Established That is the prize in units: something like 280 megabases of sequence that matters, is not coding, and is at present mostly unannotated at the level a clinician could act on.

Frontier Who collects it is unusually clear, and it is not the people doing the arguing. The gain lands on rare-disease diagnosis, where a majority of patients who reach exome sequencing get no answer; on variant-effect prediction, where the noncoding genome is where the unexplained heritability of common disease sits; and on the reference infrastructure that Precision Medicine and Neurogenetics both consume. Speculative The evolutionary-genomics side of this dispute gains nothing material from being right, which is part of why it has been the more careful side.

Handwave The condition, stated plainly: the upside is bounded not by sequencing but by the absence of an agreed definition of function, and no amount of data settles a definitional disagreement. Section 5 is about the measurement gap; section 10 is about the part that is a convention problem rather than a science problem, and the convention has been outstanding since 2012.

About 1.5% of the human genome encodes protein. The question this brief is about is what the rest of it does, and it has been contested continuously since 2012 in a way that most scientific disputes are not: both sides have published book-length arguments, both sides have the data, and the disagreement is substantially about what the word “function” picks out.

Established The two numbers that frame everything. In September 2012 the ENCODE consortium reported that “the vast majority (80.4%) of the human genome participates in at least one biochemical RNA- and/or chromatin-associated event in at least one cell type”, from 1,640 datasets across 147 cell types. Established In July 2014 a comparative-genomics analysis across 13 eutherian mammals estimated that 8.2% (7.1–9.2%) of the human genome — about 253 megabases — is presently subject to negative selection, and its authors stated that this is “around ten-fold lower than the quantity of sequence covered by the ENCODE defined elements”.

Frontier The gap is not a measurement error. It is two different questions given the same name. ENCODE measured whether a base is touched by something a cell does: transcribed, bound, modified, made accessible. The comparative analysis measured whether changing a base has historically cost its bearer offspring. Established A base can be touched without mattering, and can matter without being conserved. Frontier Both of those statements are true, both camps agree they are true, and the entire dispute is about their relative magnitudes.

Established The philosophical vocabulary is settled even though the biology is not. The causal-role sense of function asks what a thing does; the selected-effect sense asks what it was maintained for. Frontier ENCODE’s 80.4% is a causal-role number, its critics’ 8–11% is a selected-effect number, and a 2022 proposal in Genome Biology and Evolution goes further, separating sequence that persists without being maintained for anything — proposed as “spam DNA” — from sequence that lacks a maintenance function now. Speculative Adding a third category is a sign that two were not enough, not a sign that the question is close to resolution.

Frontier This brief treats the measured record on both sides, the deletion and perturbation experiments that could settle it, the definitional problem, and the misconceptions in wide circulation. It does not treat the editing of sequence already known to matter, which belongs to Genetic Engineering, nor the design of new sequence, which belongs to Synthetic Biology. Established Where this brief could not retrieve a source it says so and cites the work by name with no link.

Established Start with the anchor, because it is two-thirds right and the wrong third changes the picture. The 80.4% figure is exact and correctly attributed; the “8–9% under purifying selection” figure is the 2014 estimate of 8.2% with a 7.1–9.2% interval; and 80.4 divided by 8.2 is 9.8, so “differing tenfold” is arithmetically fair. Established What the framing gets wrong is that this is a standoff between two current numbers. It is not, and has not been since 2020.

Established ENCODE’s own successor annotation covers 7.9% of the genome, not 80%. The consortium’s Phase 3 product, published in Nature in July 2020, is a registry of 926,535 human and 339,815 mouse candidate cis-regulatory elements “covering 7.9 and 3.4% of their respective genomes”, built from 5,992 new datasets across 503 cell or tissue types. Frontier The operational product of the project that produced the 80.4% headline is a list covering under a tenth of the genome, and it is the list everybody actually uses.

Established And the constraint estimate has gone up, not down. The Zoonomia consortium’s 2023 alignment of 240 placental mammals — with 427 species in the protein-coding alignment — reports that “at least 10.7% of the human genome is evolutionarily conserved”, about 332 megabases, and identifies about 101 million significantly constrained single bases at a false-discovery rate below 0.05, which is 3.26% of the genome, 80% of it outside protein-coding exons. Established Earlier estimates ranged from 3% to 12%, and the 29-mammal alignment of 2011 resolved constraint only to about 12 bases. Frontier So the honest 2026 comparison is 7.9% against 10.7%, and the constraint figure is the larger one. Stating the dispute as 80.4 against 8.2 preserves a 2012 rhetorical moment rather than the current state of measurement.

Established A second correction to the anchor, smaller and load-bearing: the 8.2% is not a mammal-wide figure. It is constraint in the human lineage, inferred from pairwise alignments across 13 eutherian genomes. Only 2.2% of the genome has maintained constraint in both human and mouse since those lineages diverged; the remaining six percentage points are recently evolved or lineage-specific, and predominantly noncoding. Established Constrained noncoding sequence turns over with a half-life of roughly 130 million years, against over a billion years for coding sequence. Frontier Calling 8.2% “mammalian purifying selection” collapses a fourfold difference between what is constrained now and what has been constrained throughout, and that difference is the whole turnover argument.

Established The ENCODE paper itself contained the number its critics used. Its components were RNA elements 62%, histone modifications 56.1%, open chromatin 15.2%, transcription-factor binding 8.1%; and the same paper stated that comparative genomics indicates 3–8% of bases are under purifying selection. Frontier The consortium reported both figures and the press reported one.

Established The critique was immediate, specific and unusually unpleasant. A 2013 paper in Genome Biology and Evolution argued that ENCODE had adopted the causal-role definition and applied it inconsistently, that its reliance on transcriptionally permissive stem and cancer lines inflated apparent activity, that up to 90% of RNA polymerase II transcripts may be noise, and that after correcting for false positives and conservation the functional share of transcription-factor binding is nearer 0.024% than 8.5%. Frontier Its authors described the result as apophenia and offered about 9% as the defensible functional fraction. Speculative The polemic damaged the argument’s reception and did not damage its content, which is a fact about scientific culture rather than about genomes.

Established The mutational-load argument is the strongest quantitative constraint on the high estimates and it is independent of any genome annotation. A 2017 analysis argued that with a deleterious mutation rate between 4 × 10⁻¹⁰ and 10⁻⁸ per nucleotide per generation, and a maximum tolerable replacement fertility of about two children per woman, the functional fraction cannot exceed about 15%; at 80% functionality the required fertility would be 51. Frontier A 2014 review put the same argument at 70–150 mutations per generation and derived at most about 10% functional under recent models. Speculative The load argument does not say which bases are functional, and it has its own contested corrigendum and rebuttal literature; what it does is bound the answer from above without looking at a single ENCODE track.

Established The consortium’s formal reply conceded the framing and not the science. A 2014 PNAS perspective by thirty ENCODE-affiliated authors states that “the biochemically active regions cover a much larger fraction of the genome than do evolutionarily conserved regions, raising the question of whether nonconserved but biochemically active regions are truly functional”, and concludes that biochemical, evolutionary and genetic approaches are complementary and that all three are needed. Frontier That is a reasonable position and it is also a retreat from a one-number headline to a three-method framework, published eighteen months after the headline.

Established The best argument that conservation understates function is also published. A 2012 Science analysis combining 1000 Genomes variation with ENCODE activity found reduced human diversity in non-conserved regulatory elements and increased diversity in conserved elements that lack activity, implying at least 4% of additional constraint specific to the human lineage. Frontier It drew a formal Comment and a Response in Science the following year. Speculative Roughly 5% conserved across mammals plus at least 4% human-specific is the strongest version of the pro-function case built entirely from selection evidence, and it lands near 9%, not near 80%.

Established Human population data, with no cross-species comparison at all, gives a third number in the same range. A constraint map built from 76,156 human genomes in 1-kilobase non-overlapping windows found that about 3.12% of noncoding windows are as constrained as the 50th percentile of exonic regions, and about 0.05% as constrained as the 90th percentile. Frontier Three methods that share no assumptions — mammalian alignment, human population variation and mutational load — land between roughly 3% and 15%. That convergence is the strongest single fact in this brief.

Established Direct functional testing gives two numbers that look contradictory and are not. An ENCODE4 meta-analysis published in Nature Methods in March 2024 integrated 108 CRISPR-interference screens comprising more than 540,000 perturbations across 24.85 megabases, and found that 4.0% of perturbed bases — 994,400 of 24,848,100 — displayed regulatory function in their cellular context. Established A January 2025 Nature paper tested more than 680,000 candidate cis-regulatory elements by massively parallel reporter assay in three human cell types and found that about 41.7% showed regulatory activity, with promoters above 50% and enhancers between 25.8% and 42.8%. Frontier The first measures necessity where the element sits; the second measures sufficiency on a plasmid. An element can be sufficient to drive expression out of context and dispensable in place, and the tenfold gap between 4% and 42% is that distinction rather than a contradiction.

Established The same 2025 assay found that biochemical marking is a mediocre predictor of activity, which is the anchor dispute restated as a regression. Between 15% and 25% of elements with no DNase accessibility signal still exceeded negative-control thresholds, and a sequence-based deep-learning model reached a correlation of about 0.83 with measured activity against about 0.72 for biochemical models, approaching the assay’s own noise floor. Frontier Sequence beats annotation, in both directions.

Established The reference itself changed under the argument. The first gapless human assembly, published 31 March 2022, added nearly 200 million base pairs to the reference, mostly centromeric satellite DNA and segmental duplication, along with 99 likely protein-coding genes and nearly 2,000 candidates. Speculative Six per cent of the genome was missing from every ENCODE analysis before 2022, and it is the six per cent least likely to be under base-level constraint and most likely to be biochemically active, so both sides gained material.

Frontier Question one, and the one that has to be answered before the others mean anything: is pervasive transcription signal or noise? In February 2026 a group used a deep-learning transcription-initiation model on randomised versions of the complete human assembly and reported that the reversed genome yields only about 25% of the transcription start sites predicted for the forward genome, and that global mononucleotide shuffling collapses the prediction to about 0.004% of native levels. Speculative They conclude that much observed transcription cannot be attributed to background noise. Frontier Their own caveats are the reason this is a frontier claim and not a settled one: the model predicts across cell types rather than within one, measures initiation-site presence rather than transcript abundance, and the authors state explicitly that further experimental tests of the transcriptional activity of random DNA in human cells are required to confirm the model’s accuracy.

Established The experimental version of that test exists and it is the most elegant result in this brief. In 2024 a group synthesised a 101-kilobase human locus and an entirely reversed version of it — preserving base composition and repeat frequency while destroying every evolved coding and regulatory signal — and integrated both into yeast and into mouse embryonic stem cells. Established In yeast both loci were active, with about three times more transcription-start-site peaks per 100 kilobases than the genome average, and eight of eight predicted promoters drove a functional reporter. Established In mouse cells the reversed locus was silent: no accessibility, no active histone marks, no polymerase, and 0 to 4 RNA-sequencing reads across all replicates against 10 to 30 in a typical geneless 100-kilobase region. Frontier The default state of a mammalian genome is off and the default state of a yeast genome is on, which means the same observation — transcription — carries different evidential weight in the two organisms, and most of the intuitions in this debate were formed on yeast.

Frontier Question two: does conservation miss recent function, and by how much? The human-specific constraint estimate of at least 4% is the pro-function side’s best evidence, and the 130-million-year turnover half-life for constrained noncoding sequence is the sceptics’ acknowledgement of the same phenomenon. Speculative Both sides agree turnover is real; nobody has produced an agreed estimate of how much currently functional sequence is invisible to a 240-mammal alignment, and until someone does, the upper end of the plausible range is set by argument rather than by measurement.

Frontier Question three: what fraction of long noncoding RNAs do anything? A genome-scale CRISPR-interference screen of more than 16,000 long noncoding RNA loci across seven human cell lines found 499 with a growth phenotype — about 3% — and nearly 90% of those acted in only one of the seven lines. Frontier A 2026 systematic assessment of 26 genomic features reports that long noncoding RNAs give the weakest functional signatures of any gene class examined, with many “indistinguishable from intergenic regions”, against a mean effect size of 0.348 for protein-coding genes and 0.237 for short noncoding RNAs. Speculative A 3% hit rate in one assay in seven cell lines is a floor, not a ceiling, and cell-type specificity is exactly what would produce a low floor; the sceptical reading and the charitable reading of the same number are both available and neither is unreasonable.

Frontier Question four: is genome size explained by function at all? The onion test is the durable form of the challenge: Allium cepa carries about 16 billion base pairs, five times the human genome, with no corresponding complexity; genome size varies at least 7,000-fold among animals and 350-fold within vertebrates; a lungfish genome is forty times the human, a pufferfish an eighth of it. Speculative A November 2025 model argues that duplication and deletion set an equilibrium noncoding fraction determined by the product of effective population size and mutation rate, which would make genome composition a population-genetic outcome rather than an adaptive one. Handwave The onion test rules out the claim that most noncoding DNA in every genome is functional. It does not identify a single human base as junk, and treating it as though it does is the sceptical side’s characteristic overreach.

Speculative Question five: are transposable elements raw material or debris? Transposable elements make up 50–67% of the human genome, most of them degraded, and a minority have demonstrably been co-opted into regulatory roles. Frontier The 2026 transcription analysis found only about 6–10% of transcription start sites fall within repeats, and that shuffling repeat sequence barely changes their predicted initiation, which is what one would expect if repeat transcription is largely composition-driven. Handwave Exaptation is real, and no one has an estimate of its rate, so “transposable elements are a reservoir of regulatory innovation” is a true statement with no number attached to it.

Speculative Question six: can the definitional dispute be decided empirically at all? The 2022 junk-versus-spam proposal separates the question of what a sequence does now from the question of how it arrived and persisted, and argues that the junk-DNA debate only makes sense under the selected-effect definition. Handwave If that is right then no experiment settles the headline dispute, because the two camps are answering different questions correctly, and the useful work is to stop reporting a single percentage. This page treats that as the most likely resolution and the least likely to be adopted.

Frontier The binding bottleneck is fitness measurement, not sequencing. The question “does this sequence matter?” is a question about selection coefficients, and selection detectable over evolutionary time can be far below anything a laboratory can measure in a lifetime. Established A selection coefficient of 10⁻⁴ is strong enough to maintain a sequence across mammals and is invisible in a mouse colony. Handwave This is the reason the megabase-deletion experiments produce null results and the reason those nulls prove much less than they are usually taken to prove.

Established The founding deletion experiment states the bottleneck in its own discussion. Two gene deserts — 1,511 kilobases on mouse chromosome 3 and 845 kilobases on chromosome 19, jointly containing 1,243 noncoding sequences conserved between human and rodent at over 100 base pairs and 70% identity — were deleted, and homozygotes were indistinguishable from wild type across 292 and 318 offspring followed for 25 weeks, 18 plasma chemistry parameters, 13 organ types, and expression assays for 9 genes across 12 tissues. Established The only findings were reduced Prkacb in heart and reduced Rpp30 in intestine. Frontier The authors wrote that it is “possible, even likely” that the animals harbour abnormalities undetected by their assays, on some other timescale or in some other setting. Speculative That sentence is the most-ignored caveat in the entire junk-DNA literature.

Frontier The second bottleneck is redundancy, and it has been measured. Deleting ten highly conserved limb enhancers individually in mice, across 23 deletion lines, produced no noticeable change in limb morphology in any of the ten; deleting pairs near the same gene produced visible abnormalities. Established A separate study of ultraconserved elements near Arx found that pairwise deletion reduced expression and produced up to 15% lower body mass, a 62% reduction in cholinergic neurons and hippocampal abnormalities, overturning a 2007 conclusion that four such elements were dispensable. Frontier Single-element deletion is systematically underpowered against a genome built with backup, which means the standard experiment for demonstrating noncoding function is the wrong experiment.

Frontier The third is that the two available functional assays measure different things and are reported as though they measured one. Interference screens ask whether an element is necessary where it sits and return about 4%; reporter assays ask whether a sequence is sufficient on a plasmid and return about 42%. Speculative No consortium has published a matched comparison of the two on the same element set in the same cells, which is a cheap experiment and would remove one of the two largest sources of confusion in the field.

Frontier The fourth is the model organism. A large share of the genome in dispute is human-specific or primate-specific, and the assay that would test it — deletion with a fitness readout over generations — cannot be run in humans. Speculative Mouse deletion tests mouse sequence; human sequence in mouse tests it out of context, as the 2024 synthetic-locus work demonstrates by getting opposite answers in yeast and mouse for identical DNA. Handwave There is no organism in which the human-specific fraction can be tested the way the conserved fraction can, and nothing on the horizon changes that.

Speculative The fifth is definitional and it is the cheapest to fix and the least fixed. Thirteen years after the dispute began there is no agreed operational statement of what evidence licenses calling a base functional, which is why a 2026 systematic assessment of 26 genomic indicators had to begin by arguing that both biochemical activity and selection evidence are necessary. Handwave A field that must re-derive its definitions in every paper is not accumulating as fast as its data volume suggests.

Established This brief depends on three capabilities, and all three belong to other fields. Comparative genomics at species scale supplies the constraint estimates; population-scale human sequencing supplies the within-species constraint maps; and synthetic genomics supplies the ability to build and delete sequence at the scale where the question lives. Frontier Nothing on this map blocks this brief.

Established The comparative dependency has just been satisfied and the result was an upgrade. Going from 29 mammals to 240 moved constraint resolution from about 12 bases to single bases and moved the estimate from around 5% to at least 10.7%. Speculative Adding more placental mammals will improve resolution and will not touch the human-specific fraction, which is the part in dispute, so this dependency is closer to exhausted than its trajectory suggests.

Frontier The population dependency scales with cohort size and is the one still improving. A constraint map from 76,156 genomes at 1-kilobase resolution already resolves noncoding constraint; cohorts an order of magnitude larger would resolve it at the scale of individual regulatory elements. Speculative This is the dependency most likely to produce the next real number, and it depends on biobank policy rather than on technology.

Speculative The synthetic dependency is the one that would settle things and is the least mature. A 2026 review of synthetic regulatory genomics describes writing, reporter assays and saturation mutagenesis as the three routes, and names the outstanding limitation precisely: perturbational tools have not scaled to match observational genomics, and sufficiency is far better studied than necessity. Handwave Building a megabase of designed mammalian sequence and measuring fitness over many generations is the experiment this brief wants, and no published programme is attempting it. That capability is being built for other reasons in Synthetic Biology and Designer Organisms.

Frontier Traffic with the rest of this map runs outward. Genetic Engineering inherits the target list; Precision Medicine and Neurogenetics inherit the interpretation problem; Biomolecular Condensates inherits the question of whether a noncoding transcript is doing chemistry or scaffolding a compartment. Speculative Each inheritance is only as good as the functional fraction turns out to be.

Frontier The tests below rank by what a result would settle. The first is the one this brief treats as decisive and it has been technically possible for two decades and run once, badly powered.

Speculative Tier one, and the decisive experiment for this brief: megabase-scale deletion of biochemically active but non-conserved sequence, combined with saturation mutagenesis of the same regions, read out as fitness over many generations rather than as a phenotype in one. Frontier The design has to fix the three things the 2004 gene-desert deletions could not: it must target regions chosen for biochemical activity rather than for the absence of conservation, it must delete combinatorially so that redundant elements are removed together, and it must run a competitive fitness assay across generations rather than a clinical panel on one cohort. Established The precedent showed that 2.36 megabases containing 1,243 conserved noncoding sequences can be removed from a mouse with no detectable effect on morphology, reproduction, growth, longevity or 18 plasma parameters. Handwave It also showed, by its authors’ own admission, that the assay could not have detected the effect sizes that matter, which is why the experiment needs to be run again rather than cited again.

Frontier Tier two is the cheap experiment nobody has run: the same element set, both assays. Take ten thousand candidate cis-regulatory elements, test each for necessity by interference in its native locus and for sufficiency by reporter assay in the same cell line, and publish the two-by-two table. Speculative The existing numbers — about 4% by interference across 108 screens and 24.85 megabases, about 41.7% by reporter assay across 680,000 elements — come from different elements in different cells. Frontier Nobody knows whether the necessary set is a subset of the sufficient set, and that single table would discipline more loose talk than any new assay.

Established Tier three is running and has already produced the cleanest available answer to the transcription question. Synthesising a 101-kilobase locus and its sequence-reversed twin, then integrating both into two organisms, converts a correlational claim into a controlled comparison: identical base composition, identical repeat frequency, no evolved information. Established The mouse result — 0 to 4 reads from the reversed locus — is the strongest evidence in this brief that mammalian transcription is sequence-directed rather than promiscuous. Frontier Extending it from one locus to a panel, and into differentiated cells rather than stem cells, is the obvious next step and is within reach of existing methods.

Frontier Tier four is the test the 2026 modelling paper asks for on its own behalf. Its authors state that experimental measurement of the transcriptional activity of random DNA in human cells is required to confirm their model. Speculative That is a defined experiment: insert matched random, shuffled and native sequences into a fixed human locus and measure initiation directly. Speculative A four-fold difference predicted in silico either survives that comparison or it does not, and the result would bear directly on whether the pervasive-transcription evidence for widespread function is worth anything.

Frontier Tier five is combinatorial enhancer deletion at scale, because redundancy is the known confound. Ten single deletions in mice gave ten null results; pairwise deletions at the same loci gave phenotypes. Speculative A systematic programme of pairwise and higher-order deletions around a few hundred developmental genes would convert an anecdote about redundancy into a rate, and a rate is what the whole argument lacks.

Speculative Tier six is population-scale and is a natural experiment already under way. Human structural variation removes noncoding sequence constantly; with cohorts large enough, the fitness consequences of megabase deletions can be read from carrier frequencies rather than from mice. Frontier The 76,156-genome constraint map is the first instalment. Handwave Reading selection coefficients of order 10⁻⁴ from human cohorts requires numbers nobody has assembled yet, and the limiting factor is consent and governance rather than sequencing cost.

Established And the experiments that are reported as settling this question and do not. A single-element deletion with no phenotype does not show the element is junk, because redundancy is measured and pervasive. Speculative A reporter assay showing activity does not show the element is used in place. Frontier A chromatin mark does not show anything is read. Handwave Every one of those three inferences appears in high-profile literature on both sides of this dispute, and a reader assessing any claim about the functional fraction should first establish which of the three it rests on.

Frontier There is no product at the end of this page and the requirements below are for instruments that would answer a question. The one exception is clinical variant interpretation, which is a real industry waiting on this brief’s answer.

Established The writing requirement, with the scale that has actually been reached. The state of the art in mammalian synthetic regulatory genomics is a designed locus of 101 kilobases integrated at a chosen site, with a matched control differing only in sequence order. Speculative The decisive experiment in section 7 needs the same operation at 1 to 10 megabases, repeated across tens of loci, in an organism whose fitness can be measured competitively across generations. Handwave That is one to two orders of magnitude beyond what has been demonstrated, and the 2026 review of the field names exactly this gap: perturbational tools have not scaled to match observational genomics.

Established The assay requirement is a fitness readout with a resolution nobody quotes. To distinguish a functional element from a neutral one, an assay must resolve fitness differences of order 10⁻⁴; a mouse study following 292 and 318 offspring for 25 weeks resolves effects of order 10⁻¹ at best. Frontier Three orders of magnitude separate the assay that has been run from the assay that would answer the question, and that number, not the sequencing cost, is the engineering specification for this brief.

Frontier The prediction requirement is already partly met and is the most promising line. A sequence-based deep-learning model reached a correlation of about 0.83 with measured regulatory activity across 680,000 elements, against about 0.72 for models built on biochemical annotation, and approached the assay’s technical noise floor. Speculative If sequence models saturate the assay, the bottleneck moves from prediction to the assay’s own validity, which is a better problem to have. This is the seam with Generative Biomolecular Design.

Speculative The clinical requirement is a pipeline that can classify a noncoding variant, and it does not exist at diagnostic standard. The inputs are in place: a registry of 926,535 candidate elements covering 7.9% of the genome, a single-base constraint map across 240 mammals, a human constraint map at 1-kilobase resolution, and reporter measurements on 680,000 elements. Handwave Nothing joins them into an evidence framework a clinical laboratory can defend, and until something does, Precision Medicine reports most noncoding variants as of uncertain significance.

Established The nearest page on this map is Genetic Engineering, and the two divide cleanly. That brief is about editing sequence already known to matter, and its binding constraints are delivery, durability, manufacturing and reimbursement rather than biology; its central finding is that programmable targeting is close to solved while getting the machinery into the right cells is not. Frontier This page asks the prior question — which sequence matters — and its binding constraint is that nobody can measure fitness finely enough to answer it. Neither page needs the other’s answer to make progress, which is why they are separate.

Frontier Three neighbours inherit a target list and one inherits a caution. Precision Medicine inherits the noncoding variant-interpretation problem directly: the fraction of the genome a diagnostic laboratory can reason about is the fraction this brief is arguing over. Neurogenetics inherits it in its most acute form, because the heritability of neuropsychiatric traits sits overwhelmingly in noncoding sequence. Exposomics inherits the regulatory half, since environmental effects are mediated through the elements in dispute. Handwave Generative Biomolecular Design inherits the caution: a sequence model that predicts reporter activity at correlation 0.83 is predicting an assay, and the assay is the thing whose relationship to organismal function this page says is unresolved.

Speculative Two adjacencies are conceptual rather than practical. Biomolecular Condensates is the sibling case: a noncoding transcript that scaffolds a compartment would be functional in the causal-role sense and might leave no conservation signature at all, which is a route by which unconserved sequence could be doing work. Prebiotic Chemical Networks supplies the deep-time version of the same question about what sequence is for. Frontier Neither supplies a number.

Frontier One adjacency is methodological and it is the useful one. De-Extinction Technologies and Xenobiology share this subject’s core difficulty exactly: deciding which parts of a genome have to be right, and having no test that distinguishes a part that matters from a part that merely is. That is the register of Category XI and it is why this page sits here rather than in the biology categories.

Established This dispute is institutionally asymmetric and that asymmetry explains most of its dynamics. One side is a funded consortium with a data-production mandate and a publication apparatus; the other is a dispersed group of comparative genomicists and population geneticists with no shared infrastructure. Established The consortium received about $80 million over four years in its 2007 expansion alone, with $14.6 million to Stanford, $11.5 million to Yale, $10.2 million to a single commercial array supplier, $5 million to the data coordination centre and $7.26 million to technology development; its third phase produced 5,992 datasets across 503 cell or tissue types and its fourth phase was funded in February 2017. Frontier More than 2,000 publications by non-consortium researchers use the data, which is the strongest argument that the investment was sound regardless of how the 80.4% claim is judged.

Speculative The institutional failure is a definition, and it is the cheapest outstanding item in this brief. An agreed operational statement — what evidence licenses calling a base functional, reported in a fixed form — would have prevented most of thirteen years of argument, and the 2014 consortium perspective effectively proposed one by naming biochemical, evolutionary and genetic evidence as complementary. Frontier No journal requires it, no funder requires it, and papers still report a single percentage as though it were a measurement of one thing.

Frontier The publication incentive runs one way. Reporting that a noncoding element has a function is a paper; reporting that a previously claimed element does nothing is difficult to place, and a systematic deletion programme returning mostly nulls is a decade of work with a thin bibliography at the end. Speculative That is the structural reason the decisive experiment in section 7 has been technically feasible since 2004 and has been attempted essentially once.

Established The dispute has also been conducted with unusual rancour, and that has institutional consequences. The 2013 critique called the consortium’s reasoning an evolution-free gospel and its authors apophenic; the 2012 human-constraint paper drew a formal Comment and Response in Science. Speculative Rancour raises the cost of changing position publicly, and several of the people best placed to concede a point have spent a decade being unable to afford to.

Speculative The governance requirement that is coming is about cohorts, not consortia. The next real number on the functional fraction will come from human population data at a scale beyond current biobanks, which makes consent, return-of-results policy and cross-border data sharing the rate-limiting institutions. Handwave No funder has yet framed noncoding-function estimation as a reason to build such a cohort, and it would be a weak reason next to disease genetics.

Frontier The consequential ethical issue here is clinical and it is live every week. A diagnostic laboratory that cannot interpret a noncoding variant reports it as being of uncertain significance, and a family with a severe undiagnosed condition receives a non-answer. Speculative The size of that harm scales directly with the functional fraction, which is a case where an abstract definitional dispute has a measurable cost in unresolved diagnoses.

Frontier The second is about the word “junk” in public. It was coined as a technical term for sequence not maintained by selection, and it is heard as a claim that most of the genome is worthless. Speculative Both camps have exploited the ambiguity: one to make a finding sound like an overturning in press releases, the other to make a critique sound like common sense. Handwave The 2022 proposal to split junk from spam is an attempt to fix this and has not been adopted, which suggests the ambiguity is doing work for somebody.

Speculative The third concerns what gets edited. Any programme that removes noncoding sequence — in agriculture, in engineered organisms, eventually in human germline proposals — needs an answer to the question this page says is open. Frontier A deletion judged safe because the sequence is non-conserved rests on an inference that the redundancy results specifically undermine, and that reasoning belongs in the risk assessment rather than in the background. The editing side of that is Genetic Engineering; the organism side is Designer Organisms.

Frontier The fourth is about the misuse of the dispute outside science. The claim that the genome contains no junk has been recruited into arguments about design, and the claim that it is mostly junk has been recruited into arguments about human insignificance. Speculative Neither camp is responsible for its uninvited allies, and both have an obligation to state the measured ranges rather than the headline numbers, which is a low bar that press offices routinely clear from underneath.

Speculative The fifth is about research allocation. Money spent cataloguing biochemical activity is money not spent measuring whether the activity matters, and the ratio between the two over the past fifteen years is lopsided by a large factor. Handwave Nobody has computed that ratio, and computing it would be a useful and unpopular contribution.

Speculative If the functional fraction is near the low end, the consequence is a smaller and more tractable object of study than the genome appears to be. Roughly 250 to 350 megabases of functional sequence, most of it regulatory, is a target that comparative and population methods can plausibly annotate to base resolution within a few decades. Frontier That is the optimistic reading of the sceptical position, and it is rarely stated by the sceptics, who are usually arguing about rhetoric rather than about tractability.

Speculative If it is near the high end, the consequence is that a complete functional map of the human genome is not achievable by any method now on offer. Annotating two and a half billion bases of cell-type-specific, lineage-specific, redundantly encoded regulatory sequence is not a scaled-up version of what has been done; it is a different undertaking with no demonstrated method. Handwave Nobody has costed it, and the absence of a costing is itself informative about how seriously the high estimate is taken by the people who would have to do the work.

Frontier The consequence that is concrete either way is for medicine. Most of the heritability of common disease sits outside protein-coding sequence, so the ceiling on genomic medicine for common disease is set by how much of the noncoding genome can be interpreted. Speculative That ceiling is currently unknown and is a direct function of the number this brief is about, which is the strongest argument that a definitional dispute in evolutionary genomics is not merely academic.

Speculative And the outcome that is more likely than either extreme: the question dissolves rather than resolves. A plausible end state is that “what fraction of the genome is functional?” is retired as badly posed, replaced by per-element statements with an assay, a cell type and an effect size attached, and the headline percentage survives only in textbooks and press releases. Handwave That would be a real intellectual result and an invisible one: no paper announces that a question has stopped being asked, and the participants in a thirteen-year argument are not well placed to notice that it ended.

These horizons track measurement resolution and one definitional convention, because there is no engineering programme here to track:

• 10 yr: Frontier Expect population-scale constraint maps at element resolution as cohorts pass a million genomes, and expect the human-specific constraint estimate to acquire an error bar for the first time. Speculative Expect the matched necessity-versus-sufficiency table in section 7 to be published by someone, and expect it to be the most-cited result of the decade in this subject. Speculative Expect synthetic reversed-sequence experiments to extend from one locus to a panel and to settle whether mammalian transcription is sequence-directed.
• 25 yr: Speculative This is the window in which the decisive experiment plausibly runs: megabase deletion of biochemically active non-conserved sequence with competitive fitness readouts across generations, in an organism where that is possible. Speculative A result showing measurable fitness cost would move the functional fraction toward the high end and make a complete functional map a much larger undertaking than currently assumed. Speculative A null at adequate power would be the first genuinely decisive evidence for the low end in the entire dispute.
• 50 yr: Handwave On this horizon the plausible development is that the single-percentage question is retired in favour of per-element statements with assays and effect sizes attached, and the dispute is remembered as a definitional one. Speculative Precision Medicine and Neurogenetics would inherit whatever annotation exists by then, and their ceilings are set by it.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The defensible statement is about shape rather than date: a question whose answer depends on which of two definitions is used either gets a convention or gets abandoned, and this one has had neither for thirteen years. Handwave Any specific claim about what a fully annotated human genome would make possible is an assertion, and this page does not make one.

• Depends on No typed dependency edge to another brief on this map, and the absence is informative rather than an omission. This subject waits on comparative alignments, population cohorts and synthetic-genomics capacity, none of which any page here produces. Genetic Engineering is the nearest page and supplies tools rather than a prerequisite: its editing machinery is what a deletion programme would use, and its own binding constraints are delivery and reimbursement rather than annotation. Synthetic Biology develops the megabase writing capacity the decisive experiment needs and is not blocking, because that capacity is being built for other reasons.
• Requires (not on this map) Six conditions, in the order of the tokens above. The deletion. Regions chosen for biochemical activity rather than for lack of conservation, removed combinatorially so that redundant elements go together; the 2004 precedent removed 2.36 megabases containing 1,243 conserved noncoding sequences and found nothing, and said so with a caveat nobody quotes. The fitness assay. Resolution of order 10⁻⁴, against the roughly 10⁻¹ a 25-week mouse cohort achieves — three orders of magnitude, and the real specification for this brief. The comparison. One element set tested for necessity in place and sufficiency on a reporter in the same cells, which would explain the gap between about 4% and about 42% that currently confuses every discussion of how many elements are real. The writing capacity. Designed mammalian sequence at 1 to 10 megabases, against 101 kilobases demonstrated. The definition. A fixed reporting form stating what evidence licenses calling a base functional — the cheapest item on this list and the one whose absence has done the most damage. The clinical pathway. A reimbursed route for interpreting noncoding variants, without which the annotation has no user and no funder outside basic science.
• Enables A settled functional fraction would give Precision Medicine a defensible denominator for noncoding variant interpretation and would give Neurogenetics a bounded search space for the heritability that sits outside coding sequence. It would also tell Designer Organisms how much of a genome can be removed without paying for it. No typed enabling edge is claimed, because a number that is disputed by an order of magnitude enables nothing on a schedule.
• Adjacent Comparative genomics, population genetics, synthetic genomics and deep-learning sequence modelling supply every number on this page. Within this map: Genetic Engineering for the editing tools, Precision Medicine and Neurogenetics for the interpretation problem, Exposomics for the regulatory half of environmental effects, Biomolecular Condensates for the route by which an unconserved transcript could still be doing work, and De-Extinction Technologies and Xenobiology for the shared problem of deciding which parts of a genome have to be right.

Established “ENCODE showed that 80% of the genome is functional.” It showed that 80.4% of the genome participates in at least one biochemical RNA- or chromatin-associated event in at least one of 147 cell types. Established The same paper reported that comparative genomics puts 3–8% of bases under purifying selection, and the consortium’s own 2020 registry of candidate regulatory elements covers 7.9% of the genome. Frontier The headline and the annotation the consortium actually ships differ by a factor of ten.

Established “Junk DNA was disproved.” Three methods that share no assumptions — alignment across 240 mammals, constraint across 76,156 human genomes, and the mutational-load bound — put the functional fraction between roughly 3% and 15%. Frontier The claim that most of the genome is not maintained by selection is in better empirical shape in 2026 than it was in 2012, not worse.

Established “Noncoding means junk.” It does not, and this is the sceptical side’s most common misrepresentation by its opponents. Ribosomal and transfer RNA genes, microRNAs, splice sites, promoters, enhancers, insulators and origins of replication are all noncoding and all functional by any definition. Established The Zoonomia analysis finds 80% of significantly constrained bases outside protein-coding exons, which is the clearest statement that the noncoding genome contains most of the functional noncoding sequence there is.

Frontier “It was deleted and nothing happened, so it is junk.” Ten conserved limb enhancers were deleted individually in mice with no morphological change; pairwise deletions at the same loci produced abnormalities. Established Four ultraconserved elements judged dispensable in 2007 were shown in 2018 to produce up to 15% lower body mass and a 62% reduction in cholinergic neurons when deleted in pairs. Speculative A null from a single-element deletion is evidence about the assay at least as much as about the element.

Frontier “It is transcribed, so it does something.” This is the weakest inference in wide use and the evidence cuts both ways. Established A synthetic reversed 101-kilobase locus was transcriptionally silent in mouse embryonic stem cells — 0 to 4 reads across all replicates — while being promiscuously active in yeast, which argues that mammalian transcription is sequence-directed. Frontier A 2026 modelling study found native human sequence gives roughly four times more predicted initiation than a reversed genome. Speculative Both results say transcription is not random; neither says a transcript does anything, and a genome-scale interference screen found growth phenotypes for about 3% of more than 16,000 long noncoding RNA loci, nearly 90% of them in a single cell line.

Frontier “The onion test settles it.” The onion carries about 16 billion base pairs, five times the human genome; animal genome sizes span at least 7,000-fold with no relation to complexity. Established That is decisive against the claim that most noncoding DNA in every genome is functional. Handwave It identifies no human base as junk and cannot, because a comparative argument about genome size does not locate anything.

Speculative “Conservation is the gold standard, so unconserved sequence is non-functional.” Constrained noncoding sequence turns over with a half-life near 130 million years, only 2.2% of the genome has been constrained continuously in both human and mouse, and a 2012 analysis estimated at least 4% of additional human-specific constraint. Frontier Conservation is a sufficient signal and not a necessary one, and the pro-function camp is right about this even where it is wrong about the magnitude.

Speculative “The dispute is about data.” It is mostly about definitions. Handwave Both camps accept both sets of measurements; one counts a base as functional if something happens to it, the other if changing it has cost somebody offspring, and thirteen years of argument have not produced a convention because neither side’s question is the wrong one to ask.