Toward Lab-Ready AI Synthesis Plans with Protection Strategies and Route Scoring.

AI-driven synthesis planning tools are developed to automatically create synthesis routes and thereby aid chemists in making novel molecules. However, these tools frequently produce reaction trees with selectivity issues, manifested by competing reactive sites. Therefore, manual postprocessing is often required to formulate appropriate protection strategies, which prevents full automation and thus reduces the efficiency of AI-driven synthesis planning. Here, we introduce routines for identifying competing sites and formulating context-aware protection strategies by leveraging machine learning and human chemical knowledge encoded into rules and data. Our approach considers the critical interplay of cross-functional-group competition, protecting group selection, and global reaction tree structure. Moreover, the workflow supports orthogonal and multistep protections. Finally, we propose a competing sites score, which is reflective of functional group incompatibility. The score enables analyzing and reranking revised reaction trees, effectively prioritizing reaction trees with fewer selectivity issues. By evaluating on AiZynthFinder-generated reaction trees, we show that our framework substantially reduces the amount of selectivity issues and improves the quality of the synthesis routes returned to the user, with only a modest increase in computation time per target molecule. Overall, the context-aware protecting group strategy with automated selectivity control offers enhanced AI-driven synthesis planning and thus facilitates acceleration of industrial drug discovery workflows.