MolSAGE: Self-evolving Agent Experience for Efficient State-of-the-Art Drug Discovery

Building state-of-the-art (SOTA) predictive models for drug discovery requires expensive search over tools, architectures, and training strategies. Current LLM-based agents can find SOTA solutions through extensive trial and error, but repeating this search on every new task limits their efficiency. We propose MolSAGE (Self-evolving Agent Experience), a framework that accumulates and reuses experience across tasks to build SOTA drug discovery models efficiently. MolSAGE integrates cross-task experience into its search algorithm and agent workflow, using successful solutions, lessons from failures, and execution feedback to make model development more effective and efficient. We construct a diverse molecular-property benchmark in which \method accumulates experience from 25 non-antibacterial tasks and outperforms five competing agents overall on six held-out antibacterial activity prediction tasks. We then apply the SOTA solutions identified by MolSAGE to antibiotic discovery, achieving an 82\% higher hit rate than a human-led study published in Nature and identifying a compound with two- to fourfold lower minimum inhibitory concentrations than that study's human-discovered lead antibiotics. Our results show that cumulative, transferable experience enables continually improving model development and translates stronger models into more potent antibiotic candidates.