OpenAI agents analyze mouse brain data in 32 minutes, a job that would take humans months
decoder ·
OpenAI, in collaboration with PerturbAI and Scripps Research, has successfully deployed thousands of AI agents to analyze biological data from millions of mouse cells. Operating on the GPT-Rosalind platform, these agents processed a whole-brain CRISPR atlas that maps how gene changes affect various cell types within the mouse brain. The entire complex analysis was completed in just 32 minutes, whereas human experts estimate it would take several months to a year to perform manually. Each individual agent focused on a single gene perturbation, examining cellular responses and searching scientific literature for potential explanations. The system identified unexpected differences between two disease-related genes and suggested a possible link between cholesterol regulation and neurodegeneration. In blind comparisons, human experts preferred the reports generated by the distributed agents over those from a single agent 75 percent of the time. The resulting "Reasoning Atlas" is now publicly available on GitHub, accompanied by a research paper published on bioRxiv. OpenAI states that these findings are intended to help researchers prioritize which biological mechanisms should be tested in laboratory settings first.
decoder