Retrosynthesis prediction using grammar-based neural machine translation: An information-theoretic approach

Retrosynthetic prediction is one of the main challenges in chemical synthesis that requires identifying reaction pathways and precursor molecules for synthesizing a target molecule. This requires a search over the space of plausible chemical reactions that often results in complex, multi-step, branched synthesis trees for even moderately complex organic reactions. Here, we propose an approach that performs single-step retrosynthesis prediction using SMILES grammar-based representations in a neural machine translation framework. Information-theoretic analyses of such grammar-representations reveal that they are both superior and well-suited for machine learning tasks due to their underlying redundancy and high information capacity compared to purely character-based representations. We report the top-1 prediction accuracy of 43.8% (top-5 measure of 61.4%) and syntactic validity of 95.6% (top-5 measure of 91.6%) on a standard reaction dataset. Comparing our model's performance with previous work that used purely character-based SMILES representations demonstrate improved accuracy and reduced grammatically invalid predictions.