Chemically Interpretable Prediction Models for Silicon and Germanium Bond Dissociation Enthalpies.

Despite the widespread use of machine learning (ML) to predict the bond dissociation enthalpy (BDE) of bonds involving C, H, N, O, S, P, and halogens, bonds involving heavy atoms in group 14, such as Si (silicon) and Ge (germanium), remain largely understudied. In addition, many state-of-the-art models have associated limited chemical interpretability owing to their complex features and learning mechanisms. In this work, we present a highly interpretable ML model tailored to predict Si- or Ge-containing BDEs by leveraging a carefully designed set of in-house features. We compiled a comprehensive data set of 9739 unique Si-containing BDEs with density functional theory (DFT) calculation at the (u)-M06-2X/Def2-TZVP level of theory for 4679 small Si-containing molecules. A random forest (RF) model trained in a subset of these data achieved a mean absolute error (MAE) of 1.57 kcal/mol for unseen molecules. The model accurately predicts the BDE of Si-containing bioactive compounds or amino acids in less than a second, and through domain adaptation (DA), our Si-based data set enhances the prediction accuracy for Ge-containing compounds.