Molecular Optimization
Overview
Molecular optimization searches for molecules that meet goals such as stronger target binding, better drug-like properties, or improved material performance, often while preserving a starting scaffold or satisfying synthesis constraints. Methods include evolutionary search and Bayesian optimization, as well as graph, sequence, and 3D generative models—such as diffusion and language models—guided by property predictors, docking scores, or reinforcement learning. Recent work also uses tool-using agents, fragment-based edits, and closed-loop experiments or simulations to handle multiple objectives and limited evaluation budgets; methods are commonly compared on suites such as PMO, GuacaMol, TDC, and CrossDocked. A central challenge is that gains on predicted properties or standard benchmarks may not transfer to experiments or new targets, while stronger optimization can reduce diversity, synthesizability, or similarity to useful starting molecules.
Latest papers
116 papers9 papers in the last four weeks, down 50% on the four weeks before. 0.1% of all new papers.