OpenAI, PerturbAI and Scripps Research say they pointed thousands of AI agents at a huge mouse-brain gene dataset and got a map of research leads in 32 minutes. The team estimates a person working one gene at a time would need months to a year for comparable analysis and literature review.
If it holds up, it matters because labs now produce far more data than anyone can read closely. Clues that sit outside the question an experiment set out to answer tend to get missed.
How the agents split the work
The agents ran on GPT-Rosalind, OpenAI's life-sciences model. They examined a whole-brain CRISPR atlas, which records how disrupting individual genes affects different cell types across the mouse brain, drawn from millions of cells.
Each agent took one gene disruption, analyzed the cellular responses, searched the scientific literature and proposed explanations. OpenAI calls the approach distributed biological reasoning. The individual reports were merged into a "reasoning atlas" linking genes, cell effects, publications and possible mechanisms.
What turned up
One example: Tsc1 and Tsc2 are two genes that work together to control mTOR, a regulator of cell growth, metabolism and aging. Disrupting either had similar effects in several neuron groups. In the cerebellum, though, losing Tsc2 changed dozens of genes involved in neuronal communication, while losing Tsc1 changed little.
The agents also flagged a possible link between cholesterol regulation and neurodegeneration.
What is claimed and what is not
These are company-reported results, from OpenAI's own post. The team says the agents reproduced findings from human analyses and added leads, and that their ratings of novelty and scientific value matched blinded human experts. The post does not say how many experts judged, how many agent proposals were wrong, or whether any lead has been tested in the lab. The outputs are hypotheses; the post itself says they help prioritize further experiments. Also unstated here: the cost of the run and the exact agent count.
The reasoning atlas is public, so outside researchers can check the claims themselves.
The next thing to watch is lab follow-up on the Tsc2 and cholesterol leads, since the post describes them only as hypotheses for further experiments. If an agent-generated hypothesis survives a bench test, that is the real proof. Would you trust a lead that no human read first?
Sources
Reported by the WattsUpNext desk from the sources linked below. Spot an error? Tell us at corrections@wattsupnext.com.
The WattsUpNext Brief
Get stories like this in your inbox.
One email each weekday, only the topics you choose. Real news, sourced, no fluff. Unsubscribe in one click.
