OpenAI, together with the biotech company PerturbAI and Scripps Research, demonstrated how distributed AI agents can accelerate work with huge arrays of biological data.
The team used thousands of agents based on GPT-Rosalind to analyze the whole-brain CRISPR atlas — data on the effects of gene disruptions on millions of mouse brain cells. Each agent studied one gene: it analyzed cellular responses, searched for connections in the scientific literature, and proposed possible explanations.
The entire process took 32 minutes. The researchers estimate that a comparable manual analysis and literature review for a single gene could have taken from several months to a year.
What the point is
The AI did not conduct new experiments and did not prove the cause of a disease. It quickly processed already existing data, found patterns, and proposed hypotheses that scientists can now test in the laboratory.
The results were combined into a reasoning atlas — a map that links genes, cellular effects, scientific publications, and possible mechanisms. Such an atlas helps researchers avoid sifting through data manually and instead immediately see the most promising directions.
Why this matters
The main value of the project is its scale. Thousands of agents in parallel worked through data on millions of cells and cited more than 700 scientific sources. For comparison: a single agent in the same test used only 11 sources.
In a blind comparison, biologists more often chose the reports of the distributed agents — in 75% of cases. This does not mean that the AI is always right, but it shows that the system is capable of producing useful and well-substantiated research conclusions.
What they found
Among the hypotheses are unexpected differences in the action of the Tsc1 and Tsc2 genes in the cerebellum: disruption of Tsc2 changed the activity of dozens of genes, while Tsc1 changed almost none.
Another example involves the DCTN1 gene, mutations in which are associated with a rare neurodegenerative disease — Perry syndrome. Disruption of this gene increased the expression of genes involved in cholesterol synthesis. This is not an established cause of the disease, but a possible lead that requires further verification.
What comes next
The reasoning atlas is already available to researchers: it can be used to search for new hypotheses and plan experiments.
The partners view the approach as a way to find potential drug targets faster. If the method is confirmed on other data, it could make the search for new drugs more systematic. But for now this is an early stage: the AI proposes ideas, and the final conclusions remain with scientists and laboratory experiments.
