Predicting C-H activation through hydride affinity and homolytic bond dissociation energies

Predicted bond dissociation energies (BDEs) can be used to identify C-H bonds that are most likely to react in H-abstraction reactions. However, in many cases, it is not clear whether the reaction oc- curs through a radical or carbocation intermediate. Thus, the C-H hydride affinity (hydricity) may be more predictive of reactive sites than BDEs. In this paper, we introduce HAlator, a quantum chemistry (QM)-based workflow for automatic computations of C–H hydricities, that we bench- mark against 35 experimentally determined C-H hydricities in DMSO. We train the ML model on a diverse dataset of 3278 C-H sites from 740 molecules with C–H hydricities obtained using the QM-based workflow. Our ML model predicts C–H hydricities with a mean absolute error (MAE) and a root mean squared error (RMSE) of 2.30 and 3.74 kcal/mol, respectively. Furthermore, we apply our QM-based workflow and ML model to 250 hydride transfer-like reactions (H abstrac- tions: C-N, C-C, C-X, carbene insertions, oxidations, and oxidative degradation). We further ex- plore the use of ALFABET, an ML model based on BDEs, and achieve a Matthew’s correlation coefficient (MCC) between 0.20 and 0.80 across the models.