Thousands of AI agents analyze millions of brain cells in 32 minutes OpenAI

Researchers from OpenAI, PerturbAI, and Scripps Research deployed thousands of AI agents powered by GPT-Rosalind to analyze how gene knockouts affect mouse brain cells in just 32 minutes. According to the research team, an equivalent manual analysis would take human scientists several months to a year.

Each AI agent investigated an individual genetic alteration by analyzing cell responses, retrieving relevant scientific literature, and formulating hypotheses. The individual findings were then synthesized into a unified research atlas.

What it is

• Multi-agent biology. Thousands of coordinated AI agents ran in parallel on massive single-cell perturbation datasets.
• Specific discovery. The system identified a potential link between disruption in the DCTN1 gene and cholesterol metabolism inside neurons. Mutations in DCTN1 are associated with neurodegenerative diseases with symptoms similar to Parkinson's disease.
• Hypothesis generation. The findings provide concrete, testable biological hypotheses rather than raw numbers alone.
• Scale and depth. When given identical compute resources, the multi-agent collective referenced over 700 distinct scientific papers, compared to only 11 papers cited by a single agent baseline.
• Expert evaluation. In blinded evaluations, domain experts preferred the collective reports produced by the multi-agent system 75% of the time.

Research summary

Sources

• OpenAI: Scaling AI research for complex biological data