Grammars and reinforcement learning for molecule optimization
This work addresses the problem of automating molecule design for chemical property optimization, offering a novel approach that is applicable to graph structure optimization under mixed constraints, though it appears incremental in combining existing techniques.
The authors tackled automated molecule design by introducing a simpler method for generating chemically valid SMILES strings using a new context-free grammar and masking logic, and they cast molecular property optimization as a reinforcement learning problem with a Transformer model, achieving significant improvements over previous state-of-the-art baselines and enabling generation of larger molecules with fewer model steps per atom.
We seek to automate the design of molecules based on specific chemical properties. Our primary contributions are a simpler method for generating SMILES strings guaranteed to be chemically valid, using a combination of a new context-free grammar for SMILES and additional masking logic; and casting the molecular property optimization as a reinforcement learning problem, specifically best-of-batch policy gradient applied to a Transformer model architecture. This approach uses substantially fewer model steps per atom than earlier approaches, thus enabling generation of larger molecules, and beats previous state-of-the art baselines by a significant margin. Applying reinforcement learning to a combination of a custom context-free grammar with additional masking to enforce non-local constraints is applicable to any optimization of a graph structure under a mixture of local and nonlocal constraints.