Chemoenzymatic synthesis planning guided by synthetic potential scores

Abstract

Computer-aided chemoenzymatic synthesis planning integrates the advantages of enzymatic and organic reactions to design efficient hybrid synthesis routes for a target molecule. Existing tools rely on either a step-by-step strategy or a bypass strategy. Here we introduce a synthetic potential score (SPScore) to unify these two strategies. This score is developed by training a multilayer perceptron on existing reaction databases to evaluate the potential of enzymatic or organic reactions for synthesis of a molecule. We systematically evaluate the effectiveness of the SPScore in both single-step and multi-step hybrid retrosynthesis, demonstrating its strong ability to prioritize promising reaction types. In benchmarking various chemoenzymatic retrosynthesis algorithms guided by the SPScore, we find that an asynchronous search algorithm named ACERetro yields higher efficiency and robustness that can find hybrid synthesis routes to 46% more molecules compared with the state-of-the-art tool using a test dataset consisting of 1001 molecules. We then apply ACERetro to design efficient chemoenzymatic synthesis routes for 4 FDA-approved drugs. We anticipate that the application of the SPScore will provide a new avenue for computer-aided chemoenzymatic synthesis planning, thereby advancing the synthesis of functional molecules.