# OpenAI Uses GPT-Rosalind Agents to Analyze Mouse-Brain Atlas in 32 Minutes By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/37315 Published: 2026-10-22T02:06:35.000Z Updated: 2026-10-22T02:06:35.000Z Section: Technology > The atlas covers about 7.7 million cells and 1,947 disease-associated genes; proposed biological links still require laboratory testing. OpenAI, PerturbAI and Scripps Research used nearly 2,000 GPT-Rosalind agents to map an existing mouse-brain gene atlas in 32 minutes, producing leads for laboratory work that remains to be done. The resource draws on an experiment that measured how mouse-brain cells responded after scientists disrupted 1,947 genes associated with disease. It includes data from about 7.7 million cells. The Perturb-seq approach combines CRISPR gene disruption with single-nucleus RNA sequencing, which measures gene activity inside individual cell nuclei. GPT-Rosalind agents examined the recorded cellular responses and matched them with scientific literature for possible explanations. Researchers had already generated the biological measurements; the AI system worked through those results. The combined output organizes genes, observed cell effects and earlier research in a searchable atlas. The researchers estimate that human analysts reviewing comparable data one altered gene at a time would need several months to a year. The estimate is a projection, not the result of a timed trial of human analysts. In the largest workload comparison, the distributed agents cited more than 700 references, while a single-agent setup cited 11. In blinded reviews, experts chose the distributed agents’ reports 75% of the time. That result measures reviewers’ preference between reports; it does not confirm the proposed biological explanations. One research lead involved DCTN1, a gene linked to Perry syndrome. In the atlas, disrupting DCTN1 was associated with increased activity in genes involved in cholesterol production. The agents proposed cholesterol transport as a possible explanation for scientists to investigate, but the mechanism remains a hypothesis and is not an established drug target. “AI could help more researchers explore these resources and turn unexpected observations into new questions to test,” said Sarah Teichmann, chair in stem cell medicine at the University of Cambridge and founding co-chair of the Human Cell Atlas. OpenAI and PerturbAI plan to use PerturbAI’s in-vivo data for joint drug-target discovery. The effort follows OpenAI’s Tuesday, Oct. 20 acquisition of [Orchestra Bio](), which brought drug-program coordination software into its Rosalind life-sciences tools. The agent study remains a preprint, and the proposed DCTN1 mechanism still requires experimental testing. ## Links in this article - [Orchestra Bio](https://www.tokenpost.com/news/business/36360)