Physics-guided machine learning for predicting molecular SCF energies at B3LYP level. Achieves <1 second predictions with 2.8 kcal/mol MAE by embedding Hückel theory, ZPVE modeling, and solvation corrections into a 4-stage gradient boosting ensemble. Trained on QM9, generalizes to drug-like molecules.
machine-learning quantum-chemistry rdkit gradient-boosting energy-prediction dft-calculations b3lyp6
-
Updated
Jan 6, 2026 - Python