Machine Learning Prediction of High-Yield Cobalt- and Nickel-Catalyzed Borylations

Borylation reactions catalyzed by cobalt and nickel compounds occupy their important niche in synthetic organic chemistry; however, the search of parameters for high-yield reactions can be time-consuming and expensive. Recently, machine learning-based regression models were able to accurately predict reactivity yields, still when data from the literature are used, less accurate models are obtained. In this work, transforming the regression problem into a classification problem, we managed to predict high-yield cobalt- and nickel-catalyzed borylations using reaction data taken from the literature. With the Random Forest algorithm, we achieve to get the area under the receiver operating characteristics (ROC) curve mean values of 0.93 for cobalt-catalyzed reaction models and 0.86 for nickel-catalyzed reaction models. In addition, the feature importance indicates that for Co-catalyzed reactions, the characteristics of the catalyst are the most important, while in Ni-catalyzed borylations, there is a greater influence of the characteristics of the reactants and products. We think that this study may be a viable alternative to take advantage of reported reactions and could be especially useful for those laboratories that do not have the possibility to perform high-throughput experimentation to optimize their catalytic reactions.