Artificial Intelligence
OpenAI agents analyse mouse brain data in 32 minutes
OpenAI said thousands of agents, working with PerturbAI and Scripps Research, analysed a mouse brain atlas in 32 minutes, and researchers compared the output with human reports and manual estimates.
- OpenAI said it worked with PerturbAI and Scripps Research on agents that analysed biological data from mouse cells.
- OpenAI said the agents ran on GPT-Rosalind and worked through a whole-brain CRISPR atlas.
- The analysis reportedly took 32 minutes, and the team estimated that doing it by hand would have taken several months to a year.
- In blind comparisons, human experts preferred the distributed agents' reports 75 percent of the time.
- OpenAI said the agents found differences between two disease-related genes and a possible link between cholesterol regulation and neurodegeneration.
OpenAI said it worked with PerturbAI and Scripps Research on agents that analysed biological data from mouse cells using GPT-Rosalind. The system worked through a CRISPR atlas covering the whole mouse brain.
The company said the full run took 32 minutes. It estimated that a manual version of the same work would have taken several months, up to a year.
OpenAI said the agents identified unexpected differences between two disease-related genes and a possible connection between cholesterol regulation and neurodegeneration. It also said the agents searched scientific literature for explanations while doing the analysis.
In blind tests, human experts preferred the distributed agents' reports to those from a single agent 75 percent of the time. OpenAI said the resulting Reasoning Atlas was published on GitHub, alongside a paper on bioRxiv.
OpenAI said the leads could help researchers decide which mechanisms to test in the laboratory first.
Named in this story
Companies
- OpenAI
- said its agents analysed the mouse brain data
- PerturbAI
- worked with OpenAI on the analysis
Organisations
- Scripps Research
- collaborated on the project
Products and systems
- GPT-Rosalind
- ran the agents used in the analysis
- Reasoning Atlas
- was the output the team made available on GitHub
How the source tells it
The register was promotional, with OpenAI's own performance claims and a headline comparison doing most of the work.
- hype the headline contrasted a 32-minute run with work said to take months, while repeating performance claims without independent challenge
- vendor boosterism OpenAI's account of the system's speed, findings and usefulness was presented largely on its own terms