Collaborating with PerturbAI and Scripps Research, OpenAI has demonstrated how distributed AI agents can accelerate biological research. The three organisations deployed close to 2,000 AI agents to process a large mouse brain dataset, finishing the task in just 32 minutes. By comparison, the same analysis would require human researchers between several months and a year, according to the team's estimates. OpenAI shared the findings in a preprint posted to bioRxiv on Wednesday, though the work has not yet undergone peer review.
The agents operated using GPT-Rosalind, an OpenAI model specifically designed for life sciences applications. The research drew on a whole-brain CRISPR atlas created by PerturbAI, which involved using CRISPR gene-editing technology to disable individual genes in mice and observe how these modifications affected different brain cell types.
The workflow assigned each agent responsibility for a single gene modification. These agents examined cellular responses to the change, reviewed existing scientific literature, and generated hypotheses about potential mechanisms. The team subsequently consolidated all individual agent reports into a unified reasoning atlas that connected genes, cellular effects, relevant publications, and probable biological mechanisms.
OpenAI reported that the agents both confirmed earlier findings from human-led research and uncovered previously unknown leads. When experts evaluated the outputs in blinded comparisons, they favoured the multi-agent approach 75% of the time over analyses produced by a single agent.
The distributed strategy also demonstrated broader literature coverage. When handling the largest dataset, the collective agents referenced over 700 scientific papers, whereas a single agent cited only 11.
Specific discoveries
One notable finding involved the genes Tsc1 and Tsc2, both involved in regulating cell growth. Disabling Tsc2 altered dozens of genes associated with neuronal communication in the cerebellum, while disabling Tsc1 produced minimal changes. This distinction could have implications for tuberous sclerosis, a rare inherited disorder.
In another case, inactivating Dctn1—a gene associated with Perry syndrome, a rare neurological condition—increased expression of genes responsible for cholesterol synthesis. The agents proposed laboratory experiments to investigate whether this connection contributes to nerve cell deterioration.
We build cell atlases so that other scientists can make discoveries we haven't imagined yet
Sarah Teichmann, chair in stem cell medicine at the University of Cambridge
The reasoning atlas is accessible on GitHub, and researchers can access the underlying dataset through Hugging Face.
Source: The Next Web
