Accelerating the Development of Sustainable Catalytic Processes through Data Science

Abstract

Herein, we describe the optimization of reaction conditions for a nickel-catalyzed borylation using a combination of machine learning, high-throughput experimentation, and Bayesian optimization. A machine learning model, trained on a data set from Bristol Myers Squibb, was employed to predict the yields of 144 potential reaction conditions across three potential substrates for the target borylation reaction. These predictions guided a high-throughput experimentation study, which identified promising conditions for further development. The most promising condition underwent four rounds of Bayesian optimization, resulting in reaction conditions optimized for both chemical yield (>99%) and cost efficiency. The entire hit-identification and optimization process was completed in just 120 experiments over the course of 1 week by a single scientist. These optimized conditions were successfully validated on a 40 g scale to achieve 83.5% isolated yield.