OpenAI enables nearly 2,000 Agents to conduct scientific research in parallel: analyzing 8 million brain cells in 32 minutes, uncovering new clues for neurological diseases.

Beating AI News Flash: OpenAI, in collaboration with U.S. AI biotech company PerturbAI and Scripps Research, enabled 1,946 GPT-Rosalind-based Agents to analyze experimental data from approximately 8 million mouse brain cells in parallel, taking only 32 minutes. The research team estimates that if scientists were to analyze each item individually and review the literature, it could take months to a year.

These data come from previous gene-editing experiments. Scientists disrupted different genes and observed what changes occurred in brain cells. This time, AI distributed nearly 2,000 sets of gene perturbation results to different Agents, which separately analyzed cellular changes, reviewed relevant papers, and searched for connections that might have been overlooked.

One finding involves the DCTN1 gene, whose mutations are associated with a rare disease featuring Parkinsonian symptoms. AI found that after disrupting this gene, genes responsible for cholesterol synthesis in brain cells became more active instead. Combined with existing research, it proposed an explanation: perhaps it is not that cholesterol is insufficient, but that intracellular transport has gone wrong, preventing cholesterol from being delivered to where it is needed. If this hypothesis is validated, restoring normal cholesterol transport could become a new direction for protecting neurons.

AI also found that the two genes Tsc1 and Tsc2, which normally jointly regulate cell growth, have markedly different roles in the cerebellum. Disrupting Tsc2 affects dozens of genes responsible for neural signal transmission, while disrupting Tsc1 produces almost no change. This provides more specific clues for studying the genetic disease tuberous sclerosis, but the underlying mechanisms still require subsequent experimental validation.