MoLRSI: A Recursive Self-Improving AI agent for Molecular Discovery
MoLRSI’s antibiotic screen achieved a 15.9% experimental hit rate, versus 8.7% for an expert-led workflow using the same data and library.
The antibiotic workflow balanced antibacterial activity, cytotoxicity and structural novelty. MoLRSI uses results, execution feedback and persistent cross-task memory to refine workflows, model architectures and subsequent search strategies. In a separate peptide-MHC design task, it produced eight experimentally validated hits against MART-1 and SARS-CoV-1 targets, with preferential binding over tested off-targets. The authors report that Claude Science produced no hits in that comparison.
HF Daily Papers' note
The antibiotic workflow balanced antibacterial activity, cytotoxicity and structural novelty. MoLRSI uses results, execution feedback and persistent cross-task memory to refine workflows, model architectures and subsequent search strategies. In a separate peptide-MHC design task, it produced eight experimentally validated hits against MART-1 and SARS-CoV-1 targets, with preferential binding over tested off-targets. The authors report that Claude Science produced no hits in that comparison.
HF Daily Papers' note
score 4