Accurate prediction of single-point energy is essential for understanding molecular stability, electronic structure, and environment-dependent energetics, particularly for polar and charged polymeric monomers. Density functional theory (DFT) provides reliable estimates but becomes computationally prohibitive for large and chemically diverse systems. In this work, a DFT-assisted machine-learning framework is developed for predicting single-point energies of charged polymer monomers. A chemically curated monomer data set is constructed through PubChem screening and selection based on molecular stability and synthetic feasibility. Cheminformatics descriptors generated using RDKit and Mordred are combined with DFT-derived electronic descriptors, followed by correlation filtering, recursive feature elimination, and exhaustive screening to identify compact descriptor subsets. Several supervised learning models are evaluated, with Gaussian process regression yielding the best performance, with excellent agreement with DFT in both training and test sets. The resulting model is applied to extended chemical spaces, with selected predictions validated by DFT calculations. Further interpretation using atomic contribution analysis, functional-group correlations, and Shapley additive explanations provides insight into the relationship between molecular structure and energy. The framework maintains consistent accuracy across multiple dielectric environments.
Zhu et al. (Tue,) studied this question.