Scientists have identified more than 1,800 protein interactions tied to profound autism by combining gene research with artificial intelligence, potentially opening a faster path to new treatments. A team at UC San Francisco mapped these molecular connections by studying proteins produced by high-risk autism genes, then used Google DeepMind's AlphaFold AI system to predict which proteins directly interact with each other. The findings, published in the journal Science, represent a major shift from studying individual genes to understanding how proteins work together in the brain.

Why Has Autism Research Hit a Wall?

Over the past decade, geneticists have made remarkable progress identifying hundreds of genes and mutations associated with profound autism, a condition characterized by severe intellectual disability, nonverbal or minimal speech, and often serious medical complications like epilepsy. However, this genetic knowledge alone hasn't translated into effective drugs. "They kind of hit a wall," explains Nevan Krogan, director of the Quantitative Biosciences Institute at UC San Francisco, "because scientists haven't been able to turn those gene discoveries into many promising drugs and treatments".

The missing piece was understanding the proteins themselves. Genes serve as blueprints for making proteins, but researchers lacked visibility into how those proteins actually function in the brain. "What we've been missing is the mechanistic understanding of how these mutations on the genes are seemingly resulting in autism," Krogan noted. "In order to understand that, you need to go to the proteins".

How Did Researchers Map These Protein Interactions?

The research process involved several key steps that combined traditional laboratory work with cutting-edge artificial intelligence:

• Protein Selection: Researchers selected 100 proteins derived from high-risk autism genes and injected them into lab-grown cells to identify which other proteins attached to them.
• AI-Powered Analysis: They used AlphaFold, developed by Google DeepMind, to predict which proteins in the resulting clumps were directly touching each other, a process that would have taken years using traditional methods but now takes roughly an hour.
• Mutation Testing: Scientists introduced mutations found in patients with profound autism to see what changed when the proteins were "broken" in the same way they are in the condition.
• Model Validation: The team conducted experiments in both frogs and organoids, which are lab-grown tissues that model the human brain, to confirm their findings.

"Where AI is playing an important role is being predictive about who's talking to who. Sometimes they take years to figure out. Now, we can get these insights in an hour," said Nevan Krogan.

Nevan Krogan, Director of the Quantitative Biosciences Institute at UC San Francisco

In one striking example, the researchers found that mutations weakened the connection between two proteins related to which genes turn on and off in a cell. One protein then "went rogue," turning on other genes that led to neurodevelopmental defects in the lab-grown brain tissue.

What Makes This Discovery Different?

While there are hundreds of rare, high-risk autism genes acting in more than a thousand different ways, the researchers discovered something unexpected: many of these genes converge on shared biological pathways. By mapping more than 1,800 protein interactions, the team found that the proteins frequently moved along similar routes, particularly those related to early brain development, such as how synapses are constructed and which neurons develop and when.

This convergence is significant because it suggests a new therapeutic strategy. Rather than developing separate drugs for each autism gene, future treatments could potentially target these shared protein pathways. "If we look at multiple mutations and they converge on the same process, that's strong evidence that you're dealing with the things that you want to treat," explained Dr. Matthew State, a clinical psychiatrist and geneticist at UCSF.

Dr. Matthew State, a clinical psychiatrist and geneticist at UCSF

"We've been working gene-by-gene to correct mutations. And if future drugs could target these shared protein pathways, we may not need that separate therapeutic strategy for every autism gene," said Alison Singer, president of the Autism Science Foundation.

Alison Singer, President of the Autism Science Foundation

What's the Timeline for New Treatments?

While the findings are promising, researchers emphasize that translating this discovery into approved drugs will take time. The molecular map must first be converted into drug candidates, which then require testing for safety and effectiveness in clinical trials. "It takes time and a lot of persistence and focus. This paper is unlikely to lead to therapies tomorrow, but it lays an important foundation for them," State said.

However, the pace of progress may accelerate. Most human drugs are designed to target proteins, making the newly published protein interaction map an accessible entry point for pharmaceutical development. Researchers also plan to combine recent technological advances, including AI, with strategies borrowed from other fields like cancer treatment research. "When you put together the ability to borrow from successes in other fields and the development of AI, we're at an inflection point that will hopefully shorten the time cycle," State explained.

The work has already attracted significant funding. Last week, Krogan's Quantitative Biosciences Institute received a $46 million grant from the Aligning Research to Impact Autism initiative, funded by philanthropist and Google co-founder Sergey Brin, to pursue next steps in this research.

For families affected by profound autism, the research represents a meaningful shift in how scientists approach the condition. "This is the kind of scientific advance we have been waiting for and praying for," said Alison Singer, whose daughter has severe cognitive impairment. "There's still a lot of work ahead, but this paper makes the path from genetic discovery to treatment much clearer".