OpenAI Uses GPT-Rosalind Agents to Analyze Mouse-Brain Atlas in 32 Minutes

OpenAI, PerturbAI and Scripps Research used nearly 2,000 GPT-Rosalind AI agents to study an existing map of genes in the mouse brain in just 32 minutes. The work found possible leads for further research, but scientists still need to test them in a lab.

The resource is based on an experiment that measured how mouse-brain cells reacted when scientists disrupted 1,947 genes linked to diseases. It contains data from about 7.7 million cells. The Perturb-seq method combines CRISPR, which scientists use to disrupt genes, with single-nucleus RNA sequencing, which measures gene activity inside the nucleus of each cell.

The GPT-Rosalind agents studied the recorded changes in the cells and compared them with scientific papers to find possible explanations. Scientists had already collected the biological measurements, and the AI system analyzed those results. Together, the results organize genes, the changes seen in cells and earlier research into a searchable atlas.

The researchers estimate that human analysts would need several months to a year to study similar data by examining one changed gene at a time. This is an estimate, not a result from a timed test of human analysts.

In the largest comparison of workloads, the group of agents working together cited more than 700 sources, while a setup using just one agent cited 11. In reviews where experts did not know which setup had produced each report, they chose the reports from the group of agents 75% of the time. This shows which reports the reviewers preferred. It does not prove that the suggested biological explanations are correct.

One possible research lead involved DCTN1, a gene linked to Perry syndrome. In the atlas, disrupting DCTN1 was linked to increased activity in genes that help make cholesterol. The agents suggested that cholesterol transport might explain this change. Scientists still need to test this idea, and it is not yet a proven target for a drug.

“AI could help more researchers explore these resources and turn unexpected observations into new questions to test,” said Sarah Teichmann, chair in stem cell medicine at the University of Cambridge and founding co-chair of the Human Cell Atlas.

OpenAI and PerturbAI plan to use PerturbAI’s data from experiments in living animals to work together on finding new targets for drugs. This plan follows OpenAI’s acquisition of Orchestra Bio on Tuesday, Oct. 20. The acquisition added software for coordinating drug-development programs to OpenAI’s Rosalind tools for life sciences. The AI-agent study has not yet been formally published in a scientific journal and remains a preprint. The proposed explanation for how DCTN1 works still needs to be tested in experiments.