Computational prediction of C–H hydricities and their use in predicting the regioselectivity of electron-rich C–H functionalisation reactions

We present HAlator, a fully automated quantum chemistry (QM) workflow for computing C–H hydricities and explore its potential in predicting the regioselectivity of electron-rich C–H functionalisation reactions. The workflow was benchmarked against 35 experimentally determined C–H hydricities in DMSO, yielding a mean absolute error (MAE) of 4.43 kcal/mol and a root mean squared error (RMSE) of 5.45 kcal/mol. Leveraging this approach, we generated a dataset of 3278 C–H sites across 740 molecules to train a machine learning (ML) model based on CM5 atomic charge descriptors, achieving an MAE of 2.30 kcal/mol and an RMSE of 3.74 kcal/mol relative to QM-computed hydricities. The method was further applied to 250 hydride transfer-like reactions, including C–N, C–C, and C–X bond formations, carbene insertions, and oxidative transformations. Comparative analysis with ALFABET, a bond dissociation energy (BDE)-based ML model, reveals that hydricity predictions, when combined with steric accessibility, correctly identify the reactive site in eight out of ten representative reactions, surpassing BDEs in most cases. These findings highlight hydricity as a complementary and, in some cases, superior descriptor for guiding regioselectivity predictions in electron-rich C–H functionalisation. The model is made available at regioselect.org, together with a host of other reactivity predictors.