Science paper details 37,000-agent virtual biotech for drug discovery

A Stanford Medicine-linked team reported in Science that a virtual biotech of about 37,000 AI agents analyzed some 50,000 clinical trials and designed drug candidates without a physical lab. Associate professor James Zou and graduate student Harrison Zhang said agents found that drugs aimed at cell-type-specific, switch-like 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 information available before January 2025, the agents proposed a B7-H3 antibody-drug conjugate for lung cancer; months later a pharmaceutical company independently pursued the same strategy and later received FDA breakthrough therapy designation. Zou said the next step is testing new virtual-biotech findings in real laboratories.