OpenAI Unveils Mouse Brain Reasoning Atlas From GPT-Rosalind Agents

OpenAI on October 21, 2026 announced that it, PerturbAI, and Scripps Research used thousands of AI agents powered by GPT-Rosalind to analyze existing experimental data from millions of mouse brain cells in 32 minutes, combining the agents’ reports into a publicly available reasoning atlas.

Distributed Biological Reasoning

The approach, which OpenAI calls distributed biological reasoning, divided the analysis among thousands of agents examining a whole-brain CRISPR atlas, a dataset showing how gene disruptions affect different types of cells across the mouse brain. Each agent investigated a gene disruption, analyzed cellular responses, searched the scientific literature, and proposed explanations for the results. The agents’ reports were combined into a reasoning atlas linking genes, cellular effects, relevant publications, and possible biological mechanisms.

OpenAI said the research team estimated that comparable human analysis and literature review, conducted one perturbation at a time, would take months to a year. The agents reproduced findings from human analyses while uncovering additional leads, and their novelty and scientific-value assessments aligned with judgments from blinded human experts.

Sarah Teichmann, chair in Stem Cell Medicine at the University of Cambridge and founding co-chair of the Human Cell Atlas, said in OpenAI’s announcement: “We build cell atlases so that other scientists can make discoveries we haven’t imagined yet. AI could help more researchers explore these resources and turn unexpected observations into new questions to test.”

A companion preprint, Distributed AI agents enable comprehensive discovery of biological insights at scale, was posted on bioRxiv on October 16, 2026 and has not been certified by peer review. Its authors are affiliated with the Howard Hughes Medical Institute and Scripps Research, OpenAI, PerturbAI, and the University of California San Diego, with Xin Jin of Scripps Research listed as corresponding author. The competing-interest statement discloses that six of the authors are OpenAI employees, two are PerturbAI employees, and Grace Zheng and Xin Jin are co-founders of PerturbAI.

Tsc2 Divergence and a Cholesterol Lead

The agents identified an unexpected difference between Tsc1 and Tsc2, two genes that work together to control mTOR, a major regulator of cell growth, metabolism, and aging, the announcement said. Disrupting either gene had similar effects in several groups of neurons. In the cerebellum, which helps coordinate movement, Tsc2 disruption altered the activity of dozens of genes involved in neuronal communication, while Tsc1 disruption produced little change.

The announcement notes that in people with tuberous sclerosis, abnormalities in the cerebellum are strongly linked to TSC2 mutations, and that while medicines that reduce mTOR activity can treat certain brain tumors and seizures, the learning and behavioral difficulties associated with the disease remain challenging to address. The agents pinpointed a specific group of neurons and cellular changes for further investigation.

The agents also found a lead in DCTN1, a gene whose mutations cause Perry syndrome, a neurodegenerative disease with Parkinson-like symptoms. DCTN1 is involved in moving cargo inside neurons, and in the atlas, disrupting Dctn1 increased the expression of genes involved in making cholesterol. The agents connected that observation with research suggesting that problems with cholesterol transport may cause cells to make more cholesterol, raising follow-up questions about whether DCTN1 disruption changes cholesterol distribution within neurons and whether those changes could contribute to neurodegeneration. Human experts judged the hypothesis novel, consequential, and worth following up on, according to the announcement.

Tomasz Nowakowski, an associate professor of neurological surgery at UCSF, said the possible link between neuronal function and cholesterol metabolism shows how clues scattered across different fields, once connected, can offer a new way to think about disease biology.

Distributed Agents Versus a Single Agent

The team tested whether a single agent given the entire dataset could match thousands of agents working in parallel, keeping the model and resources fixed. As the workload grew, the single agent searched less of the literature, citing only 11 references at the largest workload, while the distributed agents cited more than 700. In blinded comparisons, human experts preferred the distributed-agent reports over the single-agent reports 75% of the time.

The preprint reports that, without access to the source manuscript, the agents recovered its conclusions, predicted held-out perturbation effects, and identified perturbation relationships with shared or opposing outcomes across effect strengths. The authors state that distributed agents scaled literature coverage with workload, retrieving hundreds of times more references than single-agent mode, and that blinded human experts supported the agents’ discovery rankings.

Access to the Atlas, Code, and Dataset

The reasoning atlas and the underlying Distributed Agents framework are publicly available in a GitHub repository released under an MIT license. The repository provides the agent runtime, corpus tooling, and notebooks used for the manuscript’s analyses, figures, and supplementary tables. It documents a hosted read-only corpus service with MCP endpoints that expose Findings, Claims, supporting evidence, and analysis tools; because the hosted server does not include the raw dataset files, the repository also provides instructions for running the framework locally through the Codex CLI. The repository’s latest commit, which updated the bioRxiv citation, license, and hosted-service documentation, is dated October 21, 2026.

The analyzed dataset is PerturbAI’s whole-brain in vivo CRISPR Perturb-seq atlas on Hugging Face. The dataset card describes 7.7 million cells, 7,739,370 rows totaling 205 GB, with single-nucleus profiling across 19,070 mRNAs and 8,588 sgRNAs, measuring the effects of nearly 2,000 disease-linked genes, with cell annotations drawn from the Allen Institute Whole Mouse Brain Taxonomy. The card identifies a March 2026 bioRxiv manuscript, “Genome-scale functional mapping of the mammalian whole brain with in vivo Perturb-seq,” as the atlas’s underlying publication.

OpenAI said researchers can use the atlas to compare related genes, investigate unexpected responses, and identify promising leads to pursue, and that the findings help prioritize the further experiments needed to test the proposed biological mechanisms.