Researchers at Stanford ran a pharmaceutical company entirely populated by software agents. The cancer treatment she designed was then found, identically, by a real laboratory.
James Zou, an associate professor at Stanford Medicine, and doctoral student Harrison Zhang have set up what they describe as a virtual biotechnology company: no buildings, no human employees, but 37,000 artificial intelligence agents spread over the various professions involved in drug development. The work was published on September 17, 2026 in the journal Science.
The agents first looked for what distinguishes molecules that pass the clinical stages from those that fail. They isolated genes with switch behaviour, active in certain specific cell types: candidates with this signature move 40% more often from phase 1 to phase 2, obtain marketing authorization in 48% more cases and cause 32% fewer adverse effects.
The system then designed an antibody-drug conjugate directed against the B7-H3 protein, designed to attack fibroblasts that protect lung tumors from the immune response. Validation came in an unexpected way: several months later, a major pharmaceutical company developed exactly the same approach and earned a breakthrough therapy designation from the FDA. "The science that the agents have discovered is really exciting," says Zou, who sees it as proof that these cellular characteristics can be used to design better treatments.
Read the full article on the original source
Go to the source →