Predicting Oxidation Potentials with DFT-Driven Machine Learning.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning applications

Background:

• Predicting oxidation potential (Eox) is crucial for chemical research.
• Existing methods may lack accuracy or scalability.
• A need exists for robust, machine learning-ready datasets.

Purpose of the Study:

• Introduce OxPot, a large, open-access dataset for Eox prediction.
• Establish a reliable correlation between EHOMO and Eox.
• Facilitate the development of machine learning models for Eox.

Main Methods:

• Utilized Density Functional Theory (DFT) with PBE0/cc-pVDZ for EHOMO calculations.
• Compiled a dataset of over 15,000 diverse organic molecules (OxPot).
• Performed correlation analysis and machine learning algorithm testing.

Main Results:

• Achieved a strong near-linear correlation (R²=0.977) between EHOMO and experimental Eox.
• Reported a low root-mean-square error (RMSE) of 0.064.
• Identified key molecular descriptors influencing Eox predictions via feature importance analysis.

Conclusions:

• OxPot is a valuable, ML-ready resource for accelerating Eox prediction.
• The established correlation provides a foundation for accurate predictive models.
• Computational efficiency allows for rapid screening of new molecules.