Bayesian Optimization–Guided Development of P450BM3-Catalyzed Metal–Hydride Hydrogen Atom Transfer

Metal–hydride hydrogen atom transfer (MHAT) reactions catalyzed by engineered enzymes offer a promising platform for enantioselective alkene functionalization under mild conditions. We recently reported a repurposed cytochrome P450BM3 that catalyzes enantioselective MHAT cyclization reactions using hydrosilanes as hydride sources. A practical limitation of this enzymatic transformation is the requirement for superstoichiometric amounts of hydrosilane, which limits its practicality and industrial relevance. To overcome this limitation, we applied a Bayesian optimization algorithm based on Gaussian regression to minimize hydrosilane usage while maintaining high yield and enantioselectivity. Using either PhSiH3 or MePhSiH2 as hydride source, we simultaneously optimized multiple reaction parameters, including pH, substrate concentration, surfactant loading, and hydrosilane stoichiometry. For reactions employing PhSiH3, hydrosilane consumption could be reduced from 20 equivalents to 4 equivalents without compromising yield or enantioselectivity. These optimized conditions were broadly applicable across different diene substrates. When MePhSiH2 was used, the Bayesian optimization identified distinct parameters, affording twofold higher yields than under the reported conditions. Collectively, this study confirms the power of Bayesian optimization for efficiently navigating the parameter hyperspace and provides practical conditions that enhance the applicability of enzymatic MHAT chemistry.