MIMOSA: Multi-constraint Molecule Sampling for Molecule Optimization

Molecule optimization is a fundamental task for accelerating drug discovery, with the goal of generating new valid molecules that maximize multiple properties while maintaining similarity to input molecule. Existing generative models and reinforcement learning approaches made initial success, but still face difficulties in simultaneously optimizing properties. To address such challenges, we propose MultI-constraint MOlecule SAmpling (MIMOSA) approach, sampling framework use molecule as an guess sample from target distribution. MIMOSA first pretrains two property agnostic graph neural networks (GNNs) topology substructure-type prediction, where substructure can be either atom or single ring. For each iteration, uses GNNs’ prediction employs three basic operations (add, replace, delete) generate associated weights. The weights encode constraints including constraints, upon which select promising next iteration. enables flexible encoding property- similarity-constraints efficiently satisfy various achieved up 49.1% relative improvement over best baseline terms success rate.