HypoEvolve is a multi-agent framework that uses a generational genetic algorithm to coordinate specialized LLM agents that propose, critique, and revise scientific hypotheses across generations. Evaluated on drug-repurposing hypotheses for 34 cancer types using DepMap and Open Targets as external validation measures, it outperforms six baselines, reaching a DepMap selectivity of 0.171 versus 0.115 for the strongest baseline. The gains generalize to held-out cancer types, isolating collaboration structure — not just individual model capability — as a driver of hypothesis quality.
