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

OpenAI, PerturbAI, and Scripps Research collaborated to use thousands of AI agents powered by GPT-Rosalind to analyze existing experimental data from millions of mouse brain cells in 32 minutes. The agents performed distributed biological reasoning on a whole-brain CRISPR atlas, examining gene disruptions, cellular responses, and scientific literature to create a reasoning atlas that links genes, cellular effects, relevant publications, and potential mechanisms. This approach reproduced human findings and uncovered unexpected connections, such as differences between Tsc1 and Tsc2 in the cerebellum and a possible link between cholesterol regulation and neurodegeneration. The distributed agents cited over 700 references, far more than a single agent, and human experts preferred their reports. The reasoning atlas is now available to help researchers identify new leads for diseases like Alzheimer's, Parkinson's, and tuberous sclerosis, demonstrating how AI can scale the connection of experimental data with scattered scientific knowledge to generate testable hypotheses and guide future treatments.

(Source：OpenAI)