Data Science Guided Multiobjective Optimization of a Stereoconvergent Nickel-Catalyzed Reduction of Enol Tosylates to Access Trisubstituted Alkenes

• Published in: ACS catalysis 2024-04, Vol.14 (7), p.4699-4708
• Main Authors: Romer, Natalie P., Min, Daniel S., Wang, Jason Y., Walroth, Richard C., Mack, Kyle A., Sirois, Lauren E., Gosselin, Francis, Zell, Daniel, Doyle, Abigail G., Sigman, Matthew S.
• Format: Article
• Language: English
• ISSN: 2155-5435; 2155-5435
• DOI: 10.1021/acscatal.4c00650

Herein we report a method for a stereoconvergent synthesis of trisubstituted alkenes in two steps from simple ketone starting materials. The key step is a nickel-catalyzed reduction of the corresponding enol tosylates that predominantly relies on a monophosphine ligand to direct the stereoconvergent formation of either the E- or Z-trisubstituted alkene products. Reaction optimization was accomplished using a data science workflow including monophosphine training set design, statistical modeling, and multiobjective Bayesian optimization. The optimization campaign significantly improved access to both the E- and Z-trisubstituted products in up to ∼90:10 diastereoselectivity and >90% yield. After identifying superior ligands using training set design, only 25 reactions were required for each objective (E- and Z-isomer formation) to converge on improved reaction parameters from a search space of ∼30,000 potential conditions using the EDBO+ platform. Additionally, a hierarchical machine learning model was developed to predict the stereoselectivity of untested monophosphine ligands to achieve a validation mean absolute error (MAE) of 7.1% selectivity (0.21 kcal/mol). Ultimately, we present a synergistic data science workflow leveraging the integration of training set design, statistical modeling, and Bayesian optimization, thereby expanding access to stereodefined trisubstituted alkenes.