Development and Validation of a Chemoinformatic Workflow for Predicting Reaction Yield for Pd-Catalyzed C-N Couplings with Substrate Generalizability

A machine learning-based tool that provides conditions and predicted yields for Buchwald-Hartwig couplings from a ChemDraw™ structure input is described. The tool is built on an in-house generated experimental dataset that explores a diverse network of reactant pairings. To minimize the number of experiments necessary to produce models and maximize data value, a workflow based on unsupervised machine leaning tools was created. The workflow enables the construction of models which can successfully generalize—making predictions for reactants which are not represented in the dataset.