Using thousands of AI agents to uncover new leads in biological data

Why it matters

OpenAI, PerturbAI, and Scripps Research deployed large numbers of AI agents powered by GPT-Rosalind to analyze data from a whole-brain CRISPR atlas, a dataset showing how gene disruptions affect cell types across the mouse brain. Using an approach the team calls distributed biological reasoning, each agent investigated a single gene disruption, analyzed cellular responses, searched scientific literature, and proposed explanations, with the combined reports forming a reasoning atlas linking genes, cellular effects, publications, and possible mechanisms. The team estimated comparable human analysis would take months to a year, while the agents completed the work in minutes. The agents reproduced prior human findings and surfaced additional leads, and their judgments of novelty and scientific value aligned with blinded human experts. Notable results include divergent effects of two genes that jointly regulate cell growth in the cerebellum, and a possible connection between cholesterol regulation and neurodegeneration. The reasoning atlas is publicly available, letting researchers compare genes and identify leads for further experiments. The work illustrates how AI can help scientists extract testable ideas from data too large to explore manually.