Did Scientists Explain 85% of Autism? What the New Microexon Study Actually Found

Article type: Scientific discovery explainer and primary-source claim audit

Scope: The 2026 Nature study and U.S. genetic-testing context; not individualized medical or genetic advice for children or adults

Last updated: November 2, 2026

No. Scientists have not discovered the cause of 85% of autism cases. A major study published in Nature on October 14, 2026, identified previously overlooked genetic features and linked a set of rare variants affecting them to autism. The variants were reported in about 2.5% of autistic participants in the researchers' analysis, not 85%.

The 85% figure refers to something different: the approximate proportion of autistic people for whom researchers have not identified a high-confidence genetic explanation under the study's framework. It is a description of an existing knowledge gap, not the share of autism newly explained by the discovery.

That distinction should not diminish the study. Researchers uncovered 1,514 previously unannotated microexons, developed a machine-learning model of how these tiny pieces of genetic information are regulated, and found evidence connecting rare variants to changes in the way neurons process RNA. They also tested selected predictions in laboratory-grown human neurons.

The real breakthrough is a new way to examine genetic changes that conventional analyses may overlook. The important questions are what these changes do, how convincingly they are linked to autism, and whether the discovery changes anything for autistic people or families considering genetic testing.

The findings come from Guillermo E. Parada, Hannes Bretschneider and colleagues' original study in Nature, not from the much more dramatic claim circulating on social media.

What did scientists actually discover?

The researchers brought together three advances: finding microexons that had escaped standard annotations, learning which surrounding DNA sequences regulate them, and testing whether variants in those sequences are associated with autism.

The headline figures are worth separating because they measure different things.

The microexon discovery began with extensive RNA-sequencing data, including 4,790 human tissue samples and information from 47,380 single nuclei from cortical brain regions. According to the University of Toronto's explanation of the research, the broader analysis processed more than 100 terabytes of publicly available RNA-sequencing data.

These newly cataloged microexons did not suddenly appear in human DNA. They were already part of human biology. The researchers developed ways to detect pieces that existing reference catalogs and conventional analysis methods had missed.

What are microexons, and why would something so small matter?

A microexon is an exceptionally short segment of a gene that can be included in, or left out of, an RNA message when a cell processes genetic instructions. In this research, microexons were defined as 3 to 27 DNA or RNA letters long.

To understand why that matters, it helps to distinguish DNA from the messages cells actually use.

DNA stores genetic information. When a cell needs to use a gene, it produces an RNA copy. That first RNA copy contains segments called exons and introns. During a process called RNA splicing, the cell removes introns and joins selected exons into a finished message.

Cells do not always join the same exons together. Through alternative splicing, they can include or skip certain segments, producing different RNA messages from the same gene.

A microexon is a particularly small segment whose inclusion can change the result. In some cases, that changes a protein by only a few amino acids. Those few amino acids may nevertheless alter how the protein interacts with other molecules. In neurons, such interactions help regulate communication between cells, the development of connections, and the organization of neural networks.

There is another possibility that is central to the 2026 paper: some microexons affect how much of a protein gets made at all. Including a particular microexon can introduce a premature stop signal into an RNA message. The cell may then destroy that message through a quality-control process called nonsense-mediated mRNA decay.

In other words, a microexon may change either the instructions for a protein or whether the instructions survive long enough to be used.

The new study supplied experimental evidence for this second mechanism. In cultured mouse neurons, 8 of 11 tested examples showed the expected increase in detectable microexon-containing RNA when a pathway involved in RNA decay was inhibited. In cultured human neurons, 4 of 7 tested examples responded to a different inhibitor. Those results support the proposed mechanism for selected examples; they do not establish the behavior of every newly detected microexon.

A small piece of genetic code can have a meaningful biological effect. But not every microexon change is harmful, and not every change contributes to autism. Many are part of ordinary human development and function.

How did researchers connect microexons to autism?

Finding overlooked microexons was only the first part of the study. The researchers then needed a way to predict which genetic changes might interfere with their regulation.

First, they taught a model to recognize the splicing signals

Microexons are surrounded by stretches of DNA that help determine whether they are included in an RNA message. These regions contain binding sites and other signals recognized by proteins that regulate splicing.

The team developed a machine-learning model to recognize patterns in these regions. Rather than simply reading a list of genes and assigning an autism score, the model estimated whether a sequence had the characteristics of a neuronally spliced microexon and which changes might disrupt that regulatory pattern.

The model drew on existing splicing predictors and experimental information about RNA-binding proteins. The researchers also tested its predictions against known regulatory sequences and conducted experiments involving the protein QKI, which helps control splicing.

This was a tool for predicting biological effects of DNA variation, not an AI system that could diagnose autism from a genome.

Then, they compared autistic participants with unaffected relatives

The team applied its model to whole-genome sequencing data from three large research resources: the Simons Simplex Collection, SPARK, and MSSNG.

For the central family-based statistical analysis, the paper reports 25,132 individuals from 7,391 families. That count includes research participants beyond the autistic individuals alone; it should not be mistaken for 25,132 autistic people.

The researchers focused on 369 selected microexons and looked for DNA variants predicted to alter their processing. Rare and ultrarare variants generally had larger predicted effects than common variants.

Using a family-based method that tests whether an aggregate set of variants occurs more often among affected relatives than unaffected relatives, they identified a prioritized set of 72 variants. The associated variants were reported in approximately 2.5% of autistic participants and were absent from their unaffected siblings in the comparison described by the paper.

That is meaningful evidence of association. It does not mean the researchers proved that each variant, taken alone, causes autism, or that every person with one of these variants will be autistic.

Finally, they tested selected variants in human neurons

The investigators did not stop at computational predictions. They used laboratory splicing-reporting experiments in human induced neurons and demonstrated that representative rare and ultrarare variants changed microexon splicing.

One example involved a variant in the gene MAP4K4. The experiment provided biological support for the predicted effect of a selected variant, while the larger genomic analysis provided the evidence of association with autism.

These results reinforce one another, but they answer different questions:

The evidence is stronger than a computer model making an unsupported guess. It is not yet the same as a clinically established genetic explanation for an individual.

Where did the 85%, 82% and 2.5% claims come from?

This is the part of the story most likely to be misunderstood.

The viral wording makes it sound as though scientists identified one mechanism responsible for 85% of autism. The original paper never reports that result.

The approximately 85% figure describes what remains unexplained

The paper begins by noting that high-confidence autism-associated genetic variants have been identified in only about 15% of affected individuals under the research framework it describes. It calls the remaining approximately 85% idiopathic, meaning an established genetic explanation has not been identified.

That does not mean 85% of autism is not genetic. It means an identifiable, high-confidence genetic finding has not been established for those individuals using the relevant methods and criteria. The percentage should not be treated as a universal yield for every genetic test or every clinical population.

The 82% figure has a smaller and more specific denominator

The paper's methods provide the details behind another statistic widely repeated in descriptions of the discovery.

The researchers identified 629 carriers of at least one variant in their prioritized 72-variant set. But information about previous candidate genetic findings was not available for every carrier. Equivalent annotations were unavailable for SPARK participants, leaving 162 carriers with the necessary records.

Among those 162 people:

• 133, or 82.1%, had no previously recorded candidate genetic finding.
• 29, or 17.9%, had at least one previous candidate finding.

That establishes that many of the relevant carriers had no prior recorded genetic explanation. It does not establish that the new variants now provide a confirmed genetic diagnosis for all 133 people.

Here is the comparison that changes how the 82% figure should be read

The same methods section reports that, among 7,519 individuals with the relevant annotations in the larger comparison group, 85.7% had no prior recorded candidate genetic finding.

Source: Parada et al., Methods, “Classification of superset variant carriers by previous genetic findings”. Percentages describe prior annotations, not independently confirmed causes.

The distinction is important: The 82% figure cannot, by itself, demonstrate that the new variants are especially concentrated among people whose autism was previously genetically unexplained. The broader annotated group already had a comparably high, indeed slightly higher, share with no previous finding.

This is a descriptive comparison, not a formal test of enrichment. The groups are not independent, and their compositions and available annotations require care. It would be equally wrong to claim that this simple comparison disproves the researchers' genetic findings.

Their stronger evidence lies in the family-based variant associations, the biological relevance of the affected genes, and the laboratory demonstrations of altered splicing.

The 2.5% figure is not a new diagnostic success rate

The study reports that its prioritized variants were found in approximately 2.5% of autistic participants in the analyzed cohorts. That is a reported research finding involving predicted biologically consequential variants.

It is not evidence that genetic testing can now give an established causal explanation to 2.5% more autistic people. It is also not a reliable estimate of how many autistic people in the general population carry clinically meaningful variants of this kind.

Identifying an associated variant, showing that it alters splicing, and deciding that it explains an individual person's autism are distinct steps. The paper substantially advances the first two but does not complete the third for every carrier.

Does the study mean autism is genetic?

Genetics has long been a major part of autism research. The new findings add a possible mechanism; they do not establish for the first time that genes matter.

Autism is heterogeneous. Some people have an identifiable genetic syndrome or pathogenic variant with substantial effects. For others, risk may reflect many common genetic variants of individually small effect, rare variants, genetic changes arising spontaneously rather than being inherited, and interactions with other developmental factors.

This distinction also helps explain a different percentage that sometimes appears in autism discussions: heritability. A large 2019 five-country family study in JAMA Psychiatry estimated autism heritability at approximately 80% in its population models.

That does not mean scientists can identify an autism-causing genetic variant in 80% of people. Heritability estimates describe how much variation in a trait or diagnosis across a population is statistically attributable to genetic differences under a model. They are not a count of people whose individual genetic cause has been discovered.

The same caution applies in the opposite direction. Having no identified variant does not make a person's autism non-genetic. An associated variant also does not mean genes tell the whole story of that person's development, health, personality, or support needs.

The National Institutes of Health's overview of autism describes multiple genetic and other biological contributors rather than a single universal cause.

Wasn't the connection between microexons and autism already known?

Yes. Microexons themselves are not a 2026 discovery, and neither is their possible relationship to autism. What is new is the expanded catalog and a more systematic way to connect particular regulatory DNA variants with altered microexon processing and autism status.

The research has been developing for more than a decade.

In 2014, Manuel Irimia and colleagues published a study in Cell showing that neuronal microexons were unusually conserved across evolution and that their splicing was frequently altered in a selected set of postmortem brain samples from autistic people. The study also identified regulatory proteins involved in microexon inclusion.

Subsequent experiments in animal and cellular models investigated how altered microexon processing affects neuronal development and behaviors relevant to autism research. Such experiments provide mechanistic clues, not a way to diagnose autism in mice or infer that every human case follows the same pathway.

In February 2026, Steven Dupas and colleagues published work in Nature Communications examining the effects of directly perturbing microexons during neural development, strengthening the connection to gene-expression programs relevant to autism.

The October 2026 paper then added an expanded microexon catalog, a regulatory prediction model, and family-based analysis of rare genetic variation across large human datasets.

This progression matters. The new study is more than another observation that microexons behave differently in some autistic brains. It begins to connect specific changes in DNA with a plausible molecular mechanism and human genetic association.

How strong is the evidence, and what remains uncertain?

The findings deserve serious attention for several reasons. They draw on large, established research cohorts; identify recurring signals across independent datasets; use family-based comparisons that help limit some sources of confounding; and experimentally validate selected effects on RNA splicing.

But there are clear limits.

First, the 72 variants were prioritized through aggregate statistical analyses. The results do not give every variant its own established causal status. The observed associations cannot automatically be converted into individual diagnoses.

Second, the laboratory experiments tested selected examples. Demonstrating that particular variants change RNA splicing in human induced neurons supports the mechanism, but does not demonstrate that every variant has the same effect or that the effect produces a particular trait in an individual.

Third, the 2.5% estimate comes from particular research cohorts. Recruitment, family structure, ancestry composition, variant frequencies, filtering methods, and the choice of microexons analyzed can affect how well a finding generalizes. It should not be treated as a population prevalence estimate or a clinical detection rate.

Fourth, the analysis was selective by design. Although the researchers discovered 1,514 previously unannotated human microexons, the core genetic analysis examined a narrower set of 369. Its findings should not be read as a comprehensive assessment of every possible microexon-associated variant in the human genome.

Fifth, the clinical meaning has not been established for each carrier. A molecular effect, even one supported by laboratory testing, is not automatically a diagnosis, a prediction of support needs, or a basis for treatment.

These limitations narrow the conclusion. They do not erase the finding that a newly modeled layer of genetic regulation is associated with autism in the examined families.

Does this change genetic testing for autistic children or adults?

Not immediately. The researchers developed a discovery and interpretation method, not a validated clinical autism test.

That distinction is especially important for families who have already undergone genetic testing without receiving an explanation.

A laboratory can potentially sequence a DNA region without being able to determine what a variant in that region means. The new research is relevant largely because it improves how scientists may interpret variants in and around microexons, including regions that are difficult to assess with conventional gene-centered approaches.

Different tests also look for different classes of changes:

The American College of Medical Genetics and Genomics' clinical guideline supports considering exome or genome sequencing early for children with certain congenital anomalies, developmental delay, or intellectual disability. A 2025 American Academy of Pediatrics report addresses genetic evaluation specifically for children with intellectual disability or global developmental delay. Neither recommendation should be flattened into the claim that every autistic person requires the same test.

A 2026 joint statement from the National Society of Genetic Counselors, the American College of Medical Genetics and Genomics, and the American Society of Human Genetics notes that advances in genome sequencing and collaborative research have improved the identification of likely genetic causes and may support earlier care for related medical conditions. It is broad professional context, not a clinical recommendation about the microexon variants in this study.

Autism diagnosis is a separate clinical process. As the CDC explains, there is no blood test or single medical test that diagnoses autism; clinicians evaluate developmental history and behavior.

For a fuller breakdown of testing decisions, negative results, uncertain findings, and questions to bring to appointments, see our guide to genetic testing after an autism diagnosis.

What should someone do if earlier genetic testing found nothing?

A reasonable next step is to discuss the previous result with a genetics professional, particularly if there are additional medical or developmental concerns. Useful questions include whether prior testing was limited to microarray or targeted tests, whether exome or genome sequencing is appropriate, and whether older sequence data might warrant reanalysis as scientific knowledge improves.

This study does not establish that every negative test should be repeated or that a commercial laboratory can already interpret its 72 prioritized variants for diagnosis. Reanalysis and further testing are individualized decisions, not an automatic response to a headline.

Just as important, access to communication supports, accommodations, educational services, and appropriate healthcare does not depend on identifying a genetic cause.

Could this lead to autism treatments or gene therapies?

Potentially, the research could help scientists identify biological pathways worth investigating. But no treatment or gene therapy for autism has been demonstrated by this study.

The work describes how certain DNA variants may alter RNA splicing. Turning that knowledge into an intervention would require much more: confirming the clinical relevance of particular variants, determining the effects of changing their activity, establishing safety, and demonstrating meaningful benefit in appropriate human studies.

The University of Toronto's institutional research announcement notes that researchers are exploring therapeutic approaches targeting microexon dysregulation. That is a research direction, not evidence that a treatment exists.

It is also important not to assume that a molecular mechanism implies one treatment for all autistic people. Autism encompasses diverse developmental profiles, and the question of whether an intervention is useful must include actual health needs, safety, autonomy, and the priorities of autistic people themselves.

Frequently asked questions

Are microexons found only in autistic people?

No. Microexons are a normal part of genetic regulation in humans and other animals. The study investigated differences in their regulation and certain DNA variants associated with autism. It did not discover an autism-specific piece of DNA that everybody else lacks.

Can someone have one of these genetic variants and not be autistic?

Genetic variation and autism status are not interchangeable. The paper reports that its particular prioritized variant set was absent from unaffected siblings in the family comparisons described, but that does not establish that every microexon-affecting variant causes autism or that such variants are absent from all non-autistic people. Indeed, the researchers found predicted microexon-affecting variation in more than one-third of individuals regardless of autism status.

Are these variants inherited or spontaneous?

The underlying research examines variation in families, including rare and ultrarare changes. Autism-related genetic influences can be inherited or arise de novo, meaning they are new in a child rather than inherited from either parent. The study does not support assigning one inheritance pattern to all 72 prioritized variants.

Does this mean 2.5% more autistic people can now get a genetic diagnosis?

No. The 2.5% is the researchers' reported proportion carrying prioritized variants associated with autism in the analyzed cohorts. A clinically reportable genetic diagnosis requires additional evidence about an individual variant and its relationship to that person's medical findings.

If my genetic test was negative, does that mean my autism is not genetic?

No. A negative report means that the test and interpretation process used did not identify a reportable finding. It does not exclude genetic contributions or rule out autism. Different tests cover different kinds of genetic variation, and scientific understanding changes over time.

Does this explain why autism occurs in some families but not others?

It adds a possible contributor, not a complete explanation. The model may eventually improve interpretation of some variants, but family recurrence and individual developmental outcomes depend on more than whether a person carries one of the variants highlighted in this paper.

What this discovery actually changes

The most important result is not that scientists found a single explanation for most autistic people. They did not.

They identified genetic features that standard analyses had failed to catalog, developed a way to predict how nearby DNA changes may affect them, and found statistically and experimentally supported evidence that some of those changes are relevant to autism.

The remaining work is substantial: validating more variants, determining when they make a clinically meaningful contribution, and learning whether any of this knowledge can improve healthcare.

Finding a genetic association, demonstrating a biological mechanism, and explaining an individual person's autism are three different achievements. This study meaningfully advances the first two. It does not claim to have completed the third for 85% of autistic people.

References and Further Reading

Original study and research resources

• Parada, G. E., Bretschneider, H., Li, J. D., et al. (2026), “A microexon regulatory code reveals autism-linked genetic variation,” Nature. The primary study. Results and methods support the microexon counts, variant-association findings, laboratory experiments, and the 629-carrier, 162-annotation, and 7,519-person comparisons.
• Microexon regulatory code: archived model and research resources, Zenodo. The authors' archived code and resources for researchers interested in reproducing or extending the analysis. The underlying human genomic datasets are subject to controlled-access requirements.
• University of Toronto Donnelly Centre, “Overlooked DNA regions surrounding microexons hold missing pieces of autism's genetic puzzle” (2026). Institutional explanation of the study, including the more-than-100-terabyte processing figure and discussion of future research directions. It is a summary, not a substitute for the paper's methods.

Earlier microexon and autism research

• Irimia, M., et al. (2014), “A highly conserved program of neuronal microexons is misregulated in autistic brains,” Cell. Foundational study documenting brain-related microexon regulation and differences in selected autistic brain samples, establishing that this line of research predates 2026.
• Dupas, S. J., et al. (2026), “Splice isoform-perturbation coupled to single cell transcriptome profiling reveals functions of microexons in neurogenesis and autism-linked pathways,” Nature Communications. Experiments investigating consequences of microexon perturbation during neural development.
• Bai, D., et al. (2019), “Association of Genetic and Environmental Factors With Autism in a 5-Country Cohort,” JAMA Psychiatry. Large family-based study of autism heritability. Useful for distinguishing population heritability from the percentage of people with an identified genetic diagnosis.

Clinical diagnosis and genetic testing

• Centers for Disease Control and Prevention, “Screening for Autism Spectrum Disorder”. Explains why autism is clinically diagnosed through developmental history and behavior rather than by a single medical or genetic test.
• American College of Medical Genetics and Genomics (2021), “Exome and genome sequencing for pediatric patients with congenital anomalies or intellectual disability”. Evidence-based guidance on sequencing in the specified clinical populations; not a universal recommendation for every person with isolated autism.
• American Academy of Pediatrics (2025), “Genetic Evaluation of the Child With Intellectual Disability or Global Developmental Delay”. Contemporary approach to clinical genetics when intellectual disability or global developmental delay is present.
• National Society of Genetic Counselors, ACMG, and ASHG, “Joint Statement on Autism and Genetics” (2026). General professional context on the role of genetic and genomic testing; it is not a test-specific clinical guideline.
• National Institutes of Health, “Autism”. General overview of the complex genetic and other biological contributors to autism.

Editorial currency note: Research findings and genetic-testing recommendations are current to November 2, 2026. The 72 prioritized variants are research findings; clinical interpretation, laboratory availability, and professional guidance may change as evidence develops.

Research, authorship, and updates

• Research method: Source review of the primary Nature paper, institutional research material, and selected professional and clinical guidance. This is not a systematic review.
• Author: Sherafgan Khan.

• AI assistance: AI tools assisted with research synthesis, editing, and the featured illustration. Key numerical claims were checked against the primary paper.
• Corrections: Report factual issues through Sherafy’s contact page. See Sherafy’s Editorial Standards.
• Next review due: May 2, 2027, or sooner if the paper is corrected or clinical guidance materially changes.
• Change log — November 2, 2026: Draft prepared from the supplied package and checked against the linked primary and official sources.