Distributed agents turn a mouse-brain atlas into a larger set of testable leads

Agents divided the gene-perturbation analyses, connected them with literature and assembled a reasoning atlas in 32 minutes. This is analysis of previously collected data, not a new experiment completed in half an hour. [1]

With model and resources held fixed, the distributed approach explored more literature than a single agent; blinded experts preferred its reports in 75% of comparisons, according to the report. The months-to-a-year comparison with human work is the researchers’ estimate, not a timed human trial. Proposed links remain candidates for follow-up experiments. [1]

The accompanying research repository exposes the runtime, reports and analysis notebooks. Its hosted corpus supports inspection, while some statistical work requires the underlying local data. That creates a practical opportunity for research groups to interrogate a candidate explanation before committing scarce experimental capacity. Neither the report nor the repository establishes a treatment benefit. [2]