Thousands of artificial intelligence agents analyzed experimental data from millions of mouse brain cells in 32 minutes, identifying biological connections that could guide further research into brain disease, OpenAI said.
The team estimated comparable human analysis would take months to a year. OpenAI published the findings on Oct. 21 in “Using thousands of AI agents to uncover new leads in biological data.”
OpenAI, which develops the AI technology used in the work, collaborated with PerturbAI and Scripps Research, the company said. The proposed biological explanations need further experiments to test them, it said.
The agents examined existing data showing how disruptions to thousands of genes affected cells across the mouse brain, the company said. OpenAI did not specify the period covered by the experimental data.
Each agent investigated a gene disruption, examined cell responses, searched scientific literature and proposed explanations, the company said. The team combined their work into a collection linking genes, cell changes, publications and possible biological explanations.
Human experts preferred reports from multiple agents over those from a single agent in 75% of blinded comparisons, OpenAI said. The agents also reproduced findings from human analyses and identified additional research leads, the company said.
The comparison kept the model and resources fixed, OpenAI said. At the largest workload, the distributed agents cited more than 700 references, compared with 11 for the single agent, the company said.
One lead concerned Tsc1 and Tsc2, genes that work together to control a regulator of cell growth, metabolism and aging, OpenAI said. Disrupting either gene produced similar effects in several groups of nerve cells, the company said.
The effects differed in the cerebellum, the brain region that helps coordinate movement, OpenAI said. Tsc2 disruption changed the activity of dozens of genes involved in nerve cell communication, while Tsc1 disruption produced little change, it said.
The authors linked that difference to tuberous sclerosis, saying cerebellar abnormalities in people with the disease are strongly associated with TSC2 mutations. They said the findings identified specific cells and changes for laboratory investigation.
Another lead involved DCTN1, a gene whose mutations cause Perry syndrome, a disease with Parkinson-like symptoms, OpenAI said. Disrupting the mouse gene increased activity in genes involved in making cholesterol, the company said.
The agents connected that result with research suggesting cholesterol transport problems may prompt cells to make more cholesterol, OpenAI said. The authors proposed testing whether the disruption changes cholesterol distribution inside nerve cells and whether those changes contribute to degeneration.
Human experts considered that hypothesis novel and worth pursuing, OpenAI said. The company said the combined findings are available for researchers to compare genes, investigate unexpected cell responses and choose ideas for further experiments.
