Next upDPTV HighlightsEpisode 1 · From SF Tech Week to TechCrunch Disrupt
News

OpenAI outlines distributed-agent study of mouse brain-cell data

OpenAI, PerturbAI and Scripps Research sent thousands of agents across mouse brain-cell data, producing a public reasoning atlas of research leads.

D
Oct 21, 2026 · 1 min read

OpenAI said it worked with PerturbAI and Scripps Research to deploy thousands of GPT-Rosalind agents across a whole-brain CRISPR dataset. Their output became a public reasoning atlas designed to help scientists select research leads for follow-up experiments.

The underlying bioRxiv preprint says parallel agents examined separate facets of a whole-brain Perturb-seq experiment in minutes, then combined their reports for study-wide interpretation. The paper has not been peer reviewed. Its abstract supports the reported parallel workflow and directional comparisons, but it does not support several exact performance figures in OpenAI’s announcement.

According to OpenAI’s account, each agent investigated a gene disruption, analyzed responses by cell type, searched scientific literature and proposed explanations. The public repository provides runtime and corpus tools, saved analysis inputs and notebooks associated with the manuscript. This workflow breaks a large experiment into parallel investigations and consolidates the results, instead of asking one model to interpret the entire dataset in a single pass.

The source data comes from a separate whole-brain Perturb-seq study that reports loss-of-function analysis of 1,947 disease-associated genes across more than 7.7 million mouse-brain cells. Perturb-seq combines gene disruption with single-cell RNA measurements to track how those changes affect different cell types.

The distributed-agents preprint says blinded experts preferred the distributed reports to single-agent reports and supported its rankings of proposed discoveries. Those assessments reflect expert judgment, not laboratory validation. Leads highlighted by the team—including differences involving Tsc1 and Tsc2 in the cerebellum and a proposed connection between DCTN1 and cholesterol biology—remain hypotheses for follow-up experiments.

The preprint discloses that several authors are OpenAI employees, two are PerturbAI employees and two authors are PerturbAI co-founders.

More news