OpenAI, PerturbAI, and Scripps Research used thousands of GPT-Rosalind agents to analyze millions of mouse brain cells and identify research leads.
OpenAI, PerturbAI, and Scripps Research report that thousands of AI agents analyzed existing experimental data from millions of mouse brain cells in 32 minutes. The work, published October 21, 2026, used GPT-Rosalind agents to investigate a whole-brain CRISPR atlas and connect gene-disruption results with scientific literature.
The team divided the analysis across agents, with each examining a gene disruption, interpreting cellular responses, reviewing relevant research, and proposing possible explanations. Their results were assembled into a reasoning atlas linking genes, observed cell effects, publications, and potential biological mechanisms. The atlas is available for researchers to explore.
The researchers say the system reproduced findings from human analyses and surfaced additional leads. In one example, it highlighted differences between the effects of disrupting Tsc1 and Tsc2 in cerebellar cells. It also connected Dctn1 disruption with increased expression of cholesterol-related genes, suggesting a question for follow-up experiments about cholesterol distribution in neurons. These are hypotheses for investigation, not established explanations of disease or evidence of a treatment.
The team also compared the distributed approach with a single agent using the same model and resources. At the largest workload, the single agent cited 11 references, while the distributed agents cited more than 700. In blinded evaluations, human experts preferred the distributed-agent reports 75% of the time. OpenAI says experts’ assessments of novelty and scientific value also aligned with the agents’ work.
The results illustrate a research-assistance role for AI: screening large datasets and linking observations to dispersed scientific literature so scientists can prioritize experiments. The analysis does not replace laboratory validation; the proposed mechanisms and disease connections require further testing. Researchers can use the reasoning atlas to examine related genes, cellular responses, and potential leads for follow-up work.