A New Map of 100 Autism Risk Genes Reveals Where Mutations Disrupt the Brain

Scientists at UCSF's Quantitative Biosciences Institute (QBI) have built the largest molecular map of autism to date, tracing 100 autism-linked genes through more than 1,800 protein interactions, 87% of which had never been reported before. The map, published Aug. 27 in Science under the title "A foundational autism protein interaction atlas reveals molecular convergence," reframes autism genetics from a list of separate gene hunts to a small set of shared molecular networks where mutations converge.

The atlas is the work of QBI and UCSF's Department of Psychiatry and Behavioral Sciences through the Psychiatric Cell Map Initiative (PCMI), a long-running program that uses affinity purification–mass spectrometry (AP-MS) to pull down the proteins that physically associate with a gene product inside a cell. By running the technique across 100 autism risk genes, the team captured the protein neighborhoods around each, then layered in AlphaFold, an AI tool that predicts protein shapes, to forecast where mutations might disrupt those physical interfaces. Functional studies in Xenopus (a frog model widely used in developmental biology) and lab-grown human forebrain organoids, miniature brain tissue grown from stem cells, narrowed the list to the interactions that actually matter for early brain development.

The 1,800-interaction figure is the headline number, but the structural claim is the headline idea. The risk genes do not act as 100 independent puzzles. They cluster on a smaller set of protein complexes whose disruption, the authors argue, is what the genetic signal is really pointing at. The "precision therapies" language in the UCSF and ARIA framing is downstream of that finding; the paper itself delivers a molecular atlas and a mechanistic argument about convergence, not a clinical translation.

That second phase is what a $46 million grant from Aligning Research to Impact Autism (ARIA), announced the same week, is buying. The grant funds QBI's role in ARIA's Protein-Protein Interactions Hub, one of six research hubs in ARIA's scientific ecosystem. The hub's job is to take the interaction map and run it into the next two products a drug program needs: a finer view of which complexes are actually disrupted in patient cells, and small-molecule screens that can modulate those complexes. The map is the substrate; the funding is the chemistry.

ARIA's framing, that the $46M is "one of the most significant investments in QBI's history," is the kind of institutional language that travels well in a press release and reads thin in a lab notebook. UCSF's own release leans on the same line, and the trade coverage at GEN frames the result as a foundation for "precision therapies." The paper's authors are more conservative. The atlas points toward therapies; it does not deliver one, and the distance from a disrupted protein complex to a behavior-modifying drug is measured in years and a long failure rate that this kind of program inherits from the rest of central-nervous-system drug discovery.

For researchers outside UCSF, the practical value is the shared substrate. Autism genetics has spent a decade producing ever-larger gene lists from sequencing studies. A map of where the products of those genes physically meet gives every lab with one of those risk genes a common place to look, and a common set of neighbors to test against. The Medical Xpress summary of the work highlights the same convergence: the same protein complexes keep showing up across risk genes, which is what makes the atlas a tool rather than a one-off.

The next eighteen months are the readable milestone. The ARIA hub will use the map to nominate specific disrupted complexes for follow-up, run small-molecule screens against them, and publish the next layer of the network. Watch the first ARIA Protein-Protein Interactions Hub output: if it lands a structurally validated target with tractable chemical matter, the "precision therapy" framing starts to earn itself. If the next layer just adds more interactions, the map is still useful as a community resource, and the translational language stays ahead of the data.