OpenAI, AI-native biotech company PerturbAI, and Scripps Research have published the results of a year-long collaboration: a "distributed biological reasoning" framework. Nearly 2,000 AI agents work in parallel, each assigned to one gene disruption — analyzing cellular responses, searching the scientific literature, and proposing mechanistic explanations — ultimately compiling 3,397 searchable findings into a publicly released "reasoning atlas."
Point one: speed without losing fidelity. The test used the whole-brain CRISPR atlas from Xin Jin's lab at Scripps: 7.7 million mouse brain cells, nearly 2,000 genes knocked out one by one. The agent swarm finished analysis and synthesis in 32 minutes, while a human team doing the same serial analysis and literature review would need months to a year. The underlying model, GPT-Rosalind-5.5, has a knowledge cutoff of December 2025 and had never seen the atlas's original paper, yet it fully reproduced the lab's conclusions — a genuine blind test.
Point two: blinded human experts confirmed the quality, and the system produced real new leads. In blinded evaluations by doctoral-level biologists, multi-agent reports beat single-agent ones three-to-one. The system also uncovered a link between the DCTN1 gene and cholesterol metabolism, offering a new lead for drug targets in ALS, Parkinson's, and other neurodegenerative diseases.
Source: Read the original — the above is a summarized digest for reference only.
Comments