Virtual Biotech Uses 37,000 AI Agents for Drug Discovery

Summary

Stanford Medicine researchers James Zou and Harrison Zhang have described a virtual biotechnology company staffed by 37,000 AI agents, with no human employees or physical laboratory. Built from an earlier virtual lab, the system mirrors an established biotech organization: a chief science officer agent coordinates specialized teams working in parallel across target discovery, drug design and clinical-trial planning. In a paper published in Science, the researchers report that the agents analyzed and catalogued about 50,000 clinical trials in less than a week, assigning one agent to each trial to examine safety, effectiveness and available molecular data. For trials containing single-cell gene-activity data, the agents developed scores for cell-type specificity and gene-expression bimodality. Trials with high scores in both areas had better outcomes: associated drugs were 40% more likely to progress from phase 1 to phase 2, 48% more likely to reach the market and had 32% fewer adverse events than drugs with broader activity; the pattern appeared across several disease areas. The researchers then asked the system to investigate the B7-H3 protein in lung cancer. By combining molecular, cell-communication and spatial-transcriptomic evidence, the agents proposed that B7-H3-expressing fibroblasts may suppress nearby immune cells and designed an antibody-drug conjugate to deliver chemotherapy to those cells. The design used information available before January 2025, and a private pharmaceutical company independently reached the same strategy months later; its therapy subsequently received FDA breakthrough therapy designation. Zou said the team is pursuing other targets but will rely on human researchers, physical experiments and clinical validation to determine which findings work in the real world. The study was supported by several academic and government funders, including the NIH and NSF.