Stanford's Virtual Biotech turns 37,000 AI agents into a drug-discovery factory
What's the deal? Researchers at Stanford University School of Medicine have built the "Virtual Biotech", a multi-agent AI system of around 37,000 specialised agents that emulates a full pharmaceutical company — from drug-target discovery and molecule design through to safety and clinical-trial analysis. The project, led by associate professor of biomedical data science James Zou and graduate student Harrison Zhang, grew out of Zou's earlier "Virtual Lab" of five to eight agents, which designed nanobody proteins for COVID variants that were later validated in the wet lab. A paper describing the framework, with Zhang as lead author, was published in Science on 17 September 2026.
What it found: The system's clinical-trial agents analysed around 50,000 trials in under a week and uncovered single-cell gene-activity features — target specificity and bimodality — that predict trial success: drugs directed at switch-like, cell-type-specific genes were 40% more likely to advance from phase 1 to phase 2, 48% more likely to reach market and had 32% fewer adverse events. Using only data published before January 2025, the agents autonomously designed an antibody-drug conjugate against the B7-H3 (CD276) protein for lung cancer. Months later, Merck & Co.Dealroom has a profile for this one. Try Dealroom → independently developed the same therapeutic design, which has since received US FDA breakthrough therapy designation — an external validation of the AI-designed approach.
Read more: Phys.org · VentureBeat · Science
