PerturbAI and OpenAI Launch AI Reasoning Agents for Biomedical Discovery
This week, PerturbAI, OpenAI, and their partners shared results from a year-long project to build an artificial intelligence (AI) system that can reason about scientific data. The system helps scientists understand large amounts of experimental data, connect findings with existing research, and suggest ideas that can be tested in new experiments. The team described its work in a new preprint on bioRxiv titled “Distributed AI agents enable comprehensive discovery of biological insights at scale.”
Along with scientists from PerturbAI and OpenAI, the paper's authors include researchers from Scripps Research Institute, the University of California, San Diego, and the Howard Hughes Medical Institute. They described how they used the AI system to study data from PerturbAI's whole-brain atlas, which was published earlier this year. The team created thousands of AI agents to study how cells respond to changes, search scientific papers, and connect experimental results with what scientists already know about biology. The system then brought their findings together in a reasoning atlas. This atlas connected genes, cell responses, information from scientific papers, and possible explanations for how biological processes work.
When the team tested the system on the whole-brain atlas, “thousands of parallel agents thoroughly examined the dataset in minutes.” Even without access to the original research paper, the agents “recovered its conclusions, predicted held-out perturbation effects, and identified perturbation relationships with shared or opposing outcomes across effect strengths,” the scientists wrote. In other words, the system reached the paper's conclusions, predicted effects that had been kept out of the analysis, and found relationships between changes that had similar or opposite effects at different levels of strength. The full analysis took just 32 minutes. The researchers also found that spreading the work across many agents allowed the system to search much more scientific literature as the workload grew. It found hundreds of times more references than a system using just one agent.
To check how good the agents were at explaining the results, biologists reviewed their reports. They judged whether the findings were new, whether the evidence supported them, which findings deserved further study, and how confident the system was in its conclusions. According to the preprint, in blind comparisons with reports written by single AI agents, human experts preferred the reports from the reasoning-agent system in 75% of cases, or 45 out of 60 judgments. They preferred the single-agent reports in 10 cases, or 16.7%. The remaining five judgments, or 8.3%, were ties.
The work builds on a year-long partnership that is now expanding to include the search for new drug targets. The teams will use PerturbAI's privately owned causal biological data, which show how changes can cause effects in living systems. PerturbAI and OpenAI will apply their AI reasoning systems to this data to identify possible new targets for drugs. “Biomedical experiments are generating data faster than scientists can interpret it, meaning promising new discoveries are likely to remain hidden,” said Grace Zheng, PhD, CEO and co-founder of PerturbAI. “By combining large-scale causal data generated directly in living systems with frontier AI, our partnership with OpenAI is making it possible to investigate more biological data, faster and with greater depth, and to make target discovery more systematic.”
“By combining PerturbAI's experimental data with OpenAI's reasoning capabilities, we can help scientists uncover new biological insights and identify promising leads for drug discovery,” added Joy Jiao, head of life sciences at OpenAI. “The scale, quality, and speed of PerturbAI's discovery engine make them an ideal partner, and this is just the beginning of what we can achieve together. Running nearly 2,000 agents in parallel shows how we can scale biological interpretation alongside high-throughput experimentation, helping researchers turn those findings into better medicines.”