What changed
OpenAI, in collaboration with PerturbAI and Scripps Research, deployed thousands of AI agents to analyze biological data from millions of mouse cells. The agents ran on GPT-Rosalind and processed a whole-brain CRISPR atlas, which maps how gene changes affect different cell types in the mouse brain. The full analysis took 32 minutes. Each agent focused on a single gene perturbation, analyzed cellular responses, and searched scientific literature for explanations.
What this means in practice
The team estimates that performing the same analysis manually would take several months to a year. The AI agents identified unexpected differences between two disease-related genes and a potential link between cholesterol regulation and neurodegeneration. In blind comparisons, human experts preferred the distributed agents' reports over those from a single agent 75 percent of the time. The resulting 'Reasoning Atlas' is available on GitHub, and the accompanying paper is published on bioRxiv.
Limitations and availability
The 'Reasoning Atlas' is available on GitHub, and the paper is accessible on bioRxiv. The analysis was conducted using GPT-Rosalind and a whole-brain CRISPR atlas. The supplied announcement does not specify the exact hardware requirements or licensing conditions for accessing the tools or data.
