HypoEvolve: Genetic Algorithms Coordinate Multi-Agent LLMs for Scientific Hypothesis Discovery

Forum topic · 小凯 · 2026-09-16

Summary

HypoEvolve is a research framework that uses a generational genetic algorithm to coordinate specialized LLM agents for scientific hypothesis discovery, developed by Jieyuan Liu and colleagues including Eric P. Xing and Zhen Wang (arXiv:2609.15938). The system makes agent collaboration explicit through successive updates to a hypothesis population, integrating mechanism argumentation, hypothesis reconsideration, evidence assessment, and testability evaluation. To measure hypothesis quality in a scientifically meaningful way, the authors evaluate explanations of how interventions may work via drug repurposing hypotheses, validated against external evidence. They adapt DepMap and Open Targets into complementary external metrics built on experimental, genetic, and clinical evidence. Across 34 cancer types, HypoEvolve achieved the highest scores among six baselines on both metrics, reaching a DepMap selectivity of 0.171 versus 0.115 for the strongest baseline, with gains generalizing to held-out cancer types. The work supports a vision of autonomous science where AI research teams surpass individual model capabilities.

论文概要 (Paper Overview)

Field: NLP Authors: Jieyuan Liu, Mengzhou Hu, Jefferson Chen, JungHo Kong, Pratibha Jagannatha, Yiming Gao, Dexter Pratt, Hsin-Yuan Lee, Zhiting Hu, Trey Ideker, Wei Wang, Eric P. Xing, Zhen Wang Published: 2026-09-14 arXiv: 2609.15938

Abstract (Translated)

Scientific agents contribute to hypothesis discovery by synthesizing evidence, assessing proposals, and developing new explanations. Recent systems combine scientific agents with evolutionary search through critique, comparison, and revision. However, how different forms of agent collaboration affect hypothesis quality remains an open question. Answering this question requires separating the effects of agents' scientific capabilities from those of their collaboration. A framework must therefore preserve agents' scientific roles and support rules for combining, revising, and retaining hypotheses.

Building on this view, the authors introduce HypoEvolve, which makes collaboration explicit through successive updates to a hypothesis population:

A generational genetic algorithm coordinates specialized LLM agents
It integrates mechanism argumentation, hypothesis reconsideration, evidence assessment, and testability evaluation
Each generation specifies how scientific judgments and new proposals reshape the population, making the effect of collaboration on hypothesis quality directly testable

Evaluation

The evaluation centers on scientifically meaningful hypotheses—explanations of how interventions might work. Drug repurposing links these explanations to target-level biological claims that can be checked against external evidence:

DepMap and Open Targets are adapted into complementary external metrics based on experimental, genetic, and clinical evidence
Across 34 cancer types, HypoEvolve achieved the highest score among six baselines on both metrics
DepMap selectivity: 0.171 vs. 0.115 for the strongest baseline
Gains over single-shot generation generalize to held-out cancer types

Takeaway

HypoEvolve advances a vision of autonomous science: AI research teams achieving discovery capabilities beyond individual models.

--- *Auto-collected on 2026-09-16*