Thousands of GPT-Rosalind agents analyze mouse brain CRISPR atlas in 32 minutes
OpenAI, PerturbAI and Scripps Research divided the analysis of a whole-brain CRISPR atlas among thousands of AI agents powered by GPT-Rosalind, an approach OpenAI calls distributed biological reasoning. The agents interpreted existing data from millions of mouse brain cells in 32 minutes.
Details
- Method: Each agent took one gene disruption, analyzed cellular responses, searched literature and proposed explanations. Reports were combined into a reasoning atlas linking genes, cell effects, publications and possible mechanisms.
- Speed: The team estimates comparable human analysis, one perturbation at a time, would take months to a year.
- Quality: Agents reproduced human findings; novelty and value judgments aligned with blinded human experts, who preferred distributed-agent reports over single-agent reports 75% of the time.
- Versus one agent: With model and resources fixed, at the largest workload a single agent cited 11 references versus over 700 for the distributed agents.
- Leads: An unexpected cerebellum difference between Tsc1 and Tsc2 (mTOR regulators), and a possible link between Dctn1 disruption (a Perry syndrome gene) and cholesterol regulation in neurodegeneration.
- Availability: The reasoning atlas is public on GitHub; a bioRxiv preprint accompanies the work.
What happened next
These are hypotheses, not validated findings: the post says they help prioritize lab experiments that still need to be run. Results of experimental follow-up were not reported.