Part I: On The Effects of Electron-Deficient Olefins in a Ni Catalyzed Csp3–Csp3 Cross Coupling Part II: Using Machine Learning to Predict Reaction Performance and Infer Mechanism in a Pd Catalyzed C–N Coupling

Abstract

Chapter 1 outlines a mechanistic investigation into the Ni catalyzed Negishi cross-coupling reaction of 2,2-disubstituted aziridines in the presence of Fro-DO, an electron-deficient olefin ligand. The reaction proceeds through a Ni0/II catalytic cycle with a Ni azametallacyclobutane catalyst resting state. Turnover-limiting Csp3–Csp3 reductive elimination proceeds from a spectroscopically observable NiII-dialkyl intermediate. This investigation laid the foundation for performing ligand effect studies.

Chapter 2 utilizes the key mechanistic findings of Chapter 1 to study ligand effects on the aziridine cross coupling. Using computational methods olefinic ligands were compared based on their impact on key catalytic steps. Fro-DO’s improved performance in the reaction is due to both its ability to lower the C–C reductive elimination barrier and to associate favorably to NiII, originating from a stabilizing secondary interaction between the ligand’s sulfonamide group and NiII. Design of new ligands to evaluate this proposal supports this model and led to the development of a tunable ligand framework.

Chapter 3 describes the statistical analyses of a multi-dimensional dataset gathered using High-Throughput Experimentation for a Pd catalyzed C–N coupling reaction using isoxazole fragment additives. Various machine-learning regression algorithms were evaluated for prediction. A random forest (RF) model proved optimal (R2 = 0.92 and RMSE = 7.8%). We established model generalizability of the RF model along the additive dimension along with justification for use of chemical featurization. Interpretation of the RF model suggests that reaction poisoning by the additives is related to a Pd catalyzed isoxazole decomposition pathway. Experimental support for the ML derived hypothesis highlights the utility of ML as a tool to infer mechanisms.