HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses

HypoEvolve couples a generational genetic algorithm with specialized LLM agents to generate drug-repurposing hypotheses, scoring highest against six baselines across 34 cancer types with DepMap selectivity of 0.171 versus 0.115 for the strongest baseline.

HypoEvolve is a framework that coordinates specialized LLM agents through a generational genetic algorithm in which scientific judgments and new proposals reshape an evolving hypothesis population, with agents operating under explicit collaboration roles to produce, revise, and retain hypotheses. Evaluation centers on drug repurposing, tying generated hypotheses to target-level biological claims that are assessed via external measures adapted from DepMap and Open Targets across 34 cancer types. HypoEvolve achieves the highest scores against six baselines on both measures, reaching DepMap selectivity of 0.171 versus 0.115 for the strongest baseline, and its gains over single-pass generation generalize to held-out cancer types.

• HypoEvolve coordinates specialized LLM agents via a generational genetic algorithm over an evolving hypothesis population, with scientific judgments and new proposals reshaping it under explicit collaboration roles.
• Generated hypotheses link mechanistic explanations to target-level biological claims, evaluated on drug repurposing.
• Evaluation uses external measures adapted from DepMap and Open Targets across 34 cancer types.
• HypoEvolve outperforms six baselines, achieving DepMap selectivity of 0.171 versus 0.115 for the strongest baseline.
• Gains over single-pass generation generalize to held-out cancer types.