Researchers at the University of California, San Francisco have built a molecular map of how proteins produced by autism risk genes interact with one another, work that scientists in the field are describing as a resource the search for treatments has lacked.
The team mapped 1,881 protein-protein interactions involving proteins encoded by 100 autism risk genes and published the results in the journal Science on August 27, 2026. About 87 percent of the interactions had not been reported before. The finding drawing the most attention is not the raw number but the pattern: many of those interactions converged on the same biological processes.
This is laboratory research. No patients were treated, no drug was tested, and no clinical result was produced. What changed is that researchers now have a detailed picture of machinery they previously could only infer.
The Gap Between Finding a Gene and Making a Drug
Genetics research has been productive in profound autism. Scientists have identified several hundred genes and corresponding mutations that appear in people with the condition, which involves severe intellectual disability, minimal or absent speech, frequent co-occurring conditions such as epilepsy, and a need for round-the-clock care.
Those discoveries did not translate into treatments. "But they kind of hit a wall," said Nevan Krogan, director of the Quantitative Biosciences Institute at UCSF, describing why gene identification has not produced many drug candidates.
Matthew State, a clinical psychiatrist and geneticist who chairs the Department of Psychiatry and Behavioral Sciences at UCSF, described the missing element as mechanistic. Genes are blueprints for proteins, and proteins do the actual work of brain development, but researchers could not see how those proteins operated.
The two researchers were introduced more than a decade ago by their university chancellor. State had the genes and mutations; Krogan had the technology to examine the corresponding proteins.
Convergence Is the Result That Changes the Math
The team used affinity purification mass spectrometry to systematically pull proteins out of cells along with whatever attached to them, an approach described in detail by GenomeWeb. They then used AlphaFold, the artificial intelligence system developed by Google DeepMind, to predict which proteins in each cluster were directly touching.
Krogan said the AI shortened work that previously took years, and that "where AI is playing an important role is being predictive about who's talking to who." The system is imperfect but has saved substantial time and money.
The researchers then introduced mutations found in patients with profound autism to observe what broke. In one case, mutations weakened a connection between two proteins involved in switching genes on and off. One then activated genes it should not have, producing neurodevelopmental defects.
Across the full map, many interactions ran along shared pathways tied to early brain development, including how synapses are constructed and which neurons develop and when.
"If we look at multiple mutations and they converge on the same process," State said, that is strong evidence the work has reached the biology worth treating.
Daniel Geschwind, a professor of human genetics, neurology and psychiatry at UCLA who was not involved in the research, called the map "an unprecedented resource for the field" and noted that the convergence aligns with other recent publications.
The Model Systems Deserve Precision
Understanding where this work was performed matters for interpreting it.
The interaction mapping itself was carried out in HEK293T cells, a human embryonic kidney cell line that is a standard laboratory workhorse chosen for reliability rather than biological resemblance to brain tissue. The functional follow-up, testing what happened when patient mutations were introduced, used frogs and human forebrain organoids.
Organoids are lab-grown tissues that model aspects of the developing human brain. They are three-dimensional clusters of cells that reproduce some organizational features of brain tissue, and they have become a workhorse of neurodevelopmental research. They are not brains. They lack blood supply, immune cells, sensory input, and the full complexity of a developing human nervous system.
That is not a criticism of the study. It is the standard and appropriate way to do this research. It does mean that a defect observed in an organoid is evidence about a molecular mechanism, not a demonstration of what happens in a child.
Readers should also know the funding context. Krogan's institute received a $46 million grant from the Aligning Research to Impact Autism initiative, funded by Google co-founder Sergey Brin, announced by UCSF shortly before publication.
Families Should Expect a Long Road
For families raising children with profound autism, the honest framing is that this is foundational work rather than an imminent change in care.
State said so directly in an interview with NPR, noting that translating findings into drug candidates and testing them for safety and effectiveness "takes time and a lot of persistence and focus." He described the paper as unlikely to lead to therapies soon while laying an important foundation for them.
The reason researchers consider the map valuable is practical. Most human drugs are designed to target proteins, so a detailed map of protein interactions offers pharmaceutical developers an accessible starting point in a way that a list of genes does not.
Alison Singer, president of the Autism Science Foundation and the parent of a daughter with severe cognitive impairment, described the potential shift in strategy. Researchers have been working gene by gene to correct mutations, an approach that would require a separate therapy for every autism gene. If future drugs could target shared protein pathways, she said, that separate strategy for every gene may no longer be necessary.
Nothing here changes current care, and families should be cautious about clinics or products claiming to act on these findings. Evidence-based support for profound autism continues to center on behavioral and developmental therapies, speech and occupational therapy, management of co-occurring conditions such as epilepsy and gastrointestinal problems, and educational support.
Parents with questions should raise them with their child's developmental pediatrician or neurologist rather than adjusting anything based on early-stage laboratory research.
Key Questions Answered
What did the researchers actually find? They mapped 1,881 interactions among proteins produced by 100 autism risk genes and found many converged on shared early brain development pathways.
Is this a treatment for autism? No. This is laboratory research identifying potential drug targets. No treatment has been developed, tested in people, or submitted for regulatory review.
What is a brain organoid? A lab-grown three-dimensional cluster of cells that reproduces some features of developing brain tissue. Organoids lack blood supply, immune cells and sensory input.
Was the research done in organoids? The interaction mapping used a human kidney cell line. Frogs and human forebrain organoids were used to test what happened when patient mutations were introduced.
What is profound autism? A term for autism involving severe intellectual disability, minimal or absent speech, and a need for continuous care, often alongside conditions such as epilepsy.
How long until this could help patients? Researchers involved say years. Findings must be translated into drug candidates and then tested for safety and effectiveness before any approval.
Who funded this work? The institute received a $46 million grant from the Aligning Research to Impact Autism initiative, funded by Google co-founder Sergey Brin.
