Researchers introduced PharmAgents2, an AI system designed to handle drug-discovery work from research decisions through candidate exploration. In three studies, its investigations produced findings later supported by human-run experiments: a new LpxH inhibitor scaffold, active compounds for a covalent JAK1 strategy, and a previously unknown binding target for a known bioactive molecule.
The team also introduced PharmAgents-Bench v0.1, with 121 questions and three end-to-end tasks. The authors report that PharmAgents2 scored highest on those three tasks, using expert-defined rubrics; the abstract does not give numerical scores.
Why it matters
Drug-discovery researchers may see a way to use AI to generate testable hypotheses, but the reported experimental support comes from three studies and does not establish how broadly the system will work.
