(197ax) A Universal Approach for Combining Computer-Aided Retrosynthesis and Retrobiosynthesis Tools

Enzymatic and chemical reactions have distinct synthetic spaces for designing synthetic pathways to target molecules due to the different characteristics between enzymes and chemical catalysts. A recent approach named ASKCOS (hybrid version) combines enzymatic and synthetic templates to obtain hybrid pathways, which demonstrates hybrid search can generate unique promising pathways. However, the use of template makes it only suitable for template-based synthesis planning tools. To address this limitation, we developed a Synthetic Field Score (SFScore) that can determine the appropriate synthetic field for single-step retrosynthesis of a target molecule. We hypothesized that if a compound has been reported in a chemical reaction corpus as a product, the compound is more favorable to be synthesized in the chemical synthetic field, and the same for the enzymatic reaction corpus. We trained a multilayer perceptron on a database of products extracted from an organic reaction database and an enzymatic reaction database. The model provides two scores for enzymatic and chemical synthetic fields, respectively. In the test dataset, the SFScore achieved average recall rates of 89.9%. The SFScore enables the search algorithm to discover more accessible intermediates and pathways within a time frame by avoiding unnecessary synthetic field searches, particularly for complex molecules that require long-step synthetic routes. The versatility of the SFScore allows template-free and template-based retrosynthetic tools to be combined, which leverages the strength of both approaches. In a hybrid search of 400 molecules, combining the template-free retrosynthesis tool named RetroTRAE with the template-based retrosynthesis tool named ASKCOS (enzymatic version) found hybrid pathways to 127 molecules, while entirely template-based ASKCO (hybrid version) found hybrid pathways to 202 molecules with the same terminating conditions. Although RetroTRAE's long search time prevented the pair from finding pathways for more molecules, the unique search space of the template-free tool allowed it to discover 30 molecules’ pathways that ASKCO (hybrid version) could not find.