OpenAI: Thousands of agents analyze a mouse atlas in 32 minutes

OpenAI, along with PerturbAI and Scripps Research, mobilized thousands of agents to analyze a CRISPR atlas covering the mouse brain in just 32 minutes. The reports produced, made public alongside a scientific article, were preferred by experts over those generated by a single agent. The initiative relies on GPT-Rosalind and claims to have uncovered new biological pathways.

Published Results Considered More Convincing by Experts

The reports from the distributed analysis were evaluated blindly by specialists, who preferred them to those from a single agent in 75% of cases. The corpus produced is accessible in the form of a "Reasoning Atlas" available on GitHub, accompanied by a paper submitted to bioRxiv. Among the highlighted observations are unexpected differences between two genes associated with diseases and a possible link between cholesterol regulation and neurodegeneration.

An Army of Agents on GPT-Rosalind and a Complete CRISPR Atlas

This project is the result of a partnership between OpenAI, PerturbAI, and Scripps Research, which mobilized several thousand AI agents to analyze data from millions of mouse cells. According to OpenAI, these agents utilized GPT-Rosalind and explored a CRISPR atlas encompassing the entire brain, designed to map the impact of genetic modifications on various cell types.

Execution in 32 Minutes and Hypothesis Prioritization Announced

The entire process was completed in 32 minutes. Each agent handled a genetic perturbation, measured the corresponding cellular response, and searched for explanations in the scientific literature. The team estimates that a manual equivalent would take between several months and a year. OpenAI claims that the insights gained should help laboratory teams prioritize which mechanisms to test first.