The biotech company PerturbAI and OpenAI have jointly developed an AI system designed to assist researchers in evaluating large experimental data. As the companies announced, thousands of AI agents work in parallel to interpret data, link it to scientific literature, and identify promising hypotheses for testing. Scripps Research was also involved in the work; the results were published in a joint preprint and blog post.
The background is a growing bottleneck in biomedical research: experiments generate data faster than scientists can evaluate it. According to the information, the agent system analyzed PerturbAI's whole-brain genomic dataset and summarized the findings in just 32 minutes – an estimated serial human analysis and literature search would have taken months to a year. In blinded comparisons with reports from individual AI agents, human biology experts preferred the agents' interpretations in 75 percent of cases; novelty, evidence, priority for follow-up work, and confidence were evaluated.
In March 2026, PerturbAI introduced a drug discovery platform that maps genomic cause-and-effect relationships directly in living organisms. At launch, it included a brain-wide CRISPR atlas with eight million cells, which, according to the company, is the largest causal-genomic dataset from living tissue; it has been downloaded more than 40,000 times to date. Atlases for metabolic diseases and aging have since been added.
Building on the year-long collaboration, PerturbAI and OpenAI now aim to use their AI frameworks for the discovery of new drug targets. PerturbAI CEO Grace Zheng said that novel target molecules are still largely left to chance, and investment decisions often have to be made without sufficient mechanistic evidence. The combination of large causal data from living systems with frontier AI makes it possible to investigate more biological data faster and more deeply, and to make target discovery more systematic.
Joy Jiao, Head of Life Sciences at OpenAI, said that combining PerturbAI's experimental data with OpenAI's reasoning capabilities could help generate new biological insights and identify promising approaches for drug discovery. Running nearly 2,000 agents in parallel demonstrates how biological interpretation can scale alongside high-throughput experiments. PerturbAI closed an oversubscribed pre-seed funding round in Q3 2025.
