Predicting Oxidation Potentials with DFT-Driven Machine Learning

We introduce OxPot, a comprehensive open-access data
set comprising
over 15 thousand chemically diverse organic molecules. Leveraging
the precision of DFT-derived highest occupied molecular orbital energies
(EHOMO), OxPot serves as a robust platform
for accelerating the prediction of oxidation potential (Eox). Using the PBE0 hybrid functional and cc-pVDZ basis
set, we establish a strong near-linear correlation between EHOMO and experimental Eox values, achieving an exceptional correlation coefficient
(R2) of 0.977 and a low root-mean-square
error (RMSE) of 0.064. The correlation highlights the accuracy of
OxPot as a machine learning (ML)-ready resource for Eox prediction. To further facilitate future development
of ML models, we extensively tested various algorithms and conducted
a thorough feature importance analysis. This analysis offers valuable
insights into the key molecular descriptors that influence Eox predictions, thereby enhancing model interpretability
and guiding the design of more effective predictive models. Furthermore,
the computational efficiency of the methodology ensures rapid predictions
of Eox for additional chemically similar
molecules, thereby increasing its applicability for large-scale molecular
screening and broader applications in chemical research.