Advancing Base Metal Catalysis through Data Science: Insight and Predictive Models for Ni-Catalyzed Borylation through Supervised Machine Learning
Jason M. Stevens, Jun Li, Eric M. Simmons, Steven R. Wisniewski, Stacey DiSomma, Kenneth J. Fraunhoffer, Peng Geng, Bo Hao, Erika W. Jackson
Abstract
An expansive data set containing 33 substrates, 36 unique monophosphine ligands, and two solvents was produced for the NiCl2·6H2O catalyzed aryl (pseudo)halide borylation with tetrahydroxydiboron for a total of 1632 reactions. Exploratory data analysis revealed excellent reaction performance with simple triarylphosphines (P(p-F-Ph)3 and P(p-Anis)3) and mixed aryl-alkyl phosphines (PPh2Cy), in addition to the previously established high performance with Cy-JohnPhos. The data were used to train machine learning models that predicted out of sample reaction performance with a root-mean-square error of 18.4. The important features extracted from the models identified three phosphine parameters that offered reliable reactivity thresholds for identifying optimal ligand performance. The predictive models showed reasonable performance for predicting reaction yields employing ligands not included in model training, while the important feature boundaries accurately classified the performance of 10 of the 12 external ligands examined.