Retrosynthesis aids chemists in formulating synthesis routes of molecules by proposing potential reactions and starting materials. Realistic scenarios in the pharmaceutical industry often require parallel synthesis of multiple similar compounds. The conventional approach in AI-driven synthesis planning, however, is to generate synthesis routes for each target molecule independently, thus overlooking shared pathway motifs that could reduce overall synthetic effort, cost, and time. In this work, we extend AiZynthFinder to multi-target retrosynthesis by leveraging synthesizable intermediates. In addition, we introduce the intermediate score : a novel metric for ranking and prioritizing intermediates based on their contribution to joint synthesis routes. We demonstrate the utility of our framework across three test cases, showing that our approach identifies strategic and high scoring common intermediates. Moreover, by applying the approach to more than 5,000 target molecules from the PaRoutes dataset, we verify that the extracted intermediates can be used to recover patent grouping from the data. Compared to standard AiZynthFinder, our approach results in a refined list of intermediates which are shared between more target molecules.
Picazo et al. (Fri,) studied this question.