MoLRSI is an AI agent that jointly optimizes molecular discovery workflows and model architectures while maintaining a persistent cross‑task memory to enable recursive self‑improvement. Each search iteration logs workflow performance, model refinement outcomes, and execution feedback; these signals guide subsequent searches, improving both the concrete workflows and the meta‑strategy that generates them.
In an antibiotic discovery benchmark, MoLRSI automatically designed a screening workflow that balances antibacterial activity, cytotoxicity, and structural novelty. Using the same training data and screening library, the workflow achieved a 15.9% experimental hit rate, compared with 8.7% for a prior expert‑led workflow.
For de novo peptide‑MHC binder design, MoLRSI incorporated both target binding affinity and specificity into its workflows, producing eight experimentally validated hits against MART‑1 and SARS‑CoV‑1 pMHCs with preferential binding over off‑targets, while Claude Science failed to generate any viable candidates.
These findings demonstrate that experience‑driven workflow optimization can address competing objectives in molecular discovery and continuously improve the search strategies employed in future campaigns.
Review