HypoEvolve: Genetic Algorithms Orchestrate Multi-Agent LLMs for Cancer Drug Repurposing
Jieyuan Liu · hf · 2026-09-17
HypoEvolve is a hypothesis-discovery framework that uses a generational genetic algorithm to coordinate specialized LLM agents, making the effect of collaboration styles on hypothesis quality directly testable.
Method
- Collaboration is made explicit through successive updates to a hypothesis population; each generation specifies how scientific judgments and new proposals reshape the population, with agents integrating mechanistic arguments, reconsidering assumptions, and assessing evidence and testability
- Evaluation targets scientifically meaningful hypotheses: drug repurposing links explanations to target-level biological claims, measured via DepMap and Open Targets adapted into complementary external measures grounded in experimental, genetic, and clinical evidence
Results
- Across 34 cancer types, HypoEvolve beats six baselines on both measures: DepMap selectivity reaches 0.171 vs 0.115 for the strongest baseline
- Gains over single-pass generation generalize to held-out cancer types
The authors frame this as a step toward autonomous science, where AI research teams achieve discovery capacity beyond individual models.
More from Research
- New DSPy 'Jev' engine optimizes programs by splitting output fields, not modules — dbreunig · 2026-09-17
- Zeno Robotics unveils Zeno-1, a 3B foundation model for decentralized multi-robot collaboration — jiqizhixin · 2026-09-17
- New curvature-conditioned multiscale momentum optimizer significantly accelerates Muon LLM pretraining — YouJiacheng · 2026-09-17
- White paper reframes the HuggingFace incident via LLM semantic field miscalibration — LopsidedLevel9009 · 2026-09-17
- ModAR predicts robot futures one modality at a time; 30M model beats 6B video model — k7agar · 2026-09-17
- USTC and Alibaba's CEDAR cuts demand forecast error 57% by decoupling decisions from external shocks — 量子位 · 2026-09-17