Quantifying Base-Dependent Lewis Acidity in Substituted Boranes through DFT and Machine Learning

The dependence of Lewis acidity on the interacting base is established. Here, density functional theory calculations and machine-learning analysis quantify substituent effects and base dependence within a combinatorial library of 816 trisubstituted boranes constructed from 16 substituents. Interactions with five probe bases (H2O, NH3, C5H5N, PMe3, and Cl–) yield 3927 calculated adduct formation free energies in dichloromethane. Electron-withdrawing substituents generally enhance borane Lewis acidity, whereas electron-donating substituents attenuate it. The magnitude of these effects varies with the base, producing changes and occasional reversals in relative acidity rankings. PMe3 shows the strongest dependence on substituent identity among the investigated bases. Weak linear correlations between isolated-borane global hardness and adduct formation free energy indicate limited predictive power for this single descriptor. Random forest models describe the computed trends, while SHAP analysis identifies the summed Hammett σp parameter, the borane–base frontier orbital energy gap, and borane electrophilicity as the most influential descriptors in the model. These results provide a quantitative map of substituent effects and the transferability of relative borane acidity rankings across the investigated bases.