Predicting the Redox Potentials of Phenazine Derivatives Using DFT-Assisted Machine Learning.

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Redox flow batteries (RFBs) are crucial for energy storage, but require novel, efficient materials.
• Phenazine derivatives are potential candidates, but their synthesis and testing are resource-intensive.

Purpose of the Study:

• To develop accurate machine learning models for predicting redox potentials of phenazine derivatives.
• To accelerate the screening of phenazine derivatives for energy storage applications.

Main Methods:

• Investigated four machine learning models using a dataset of 151 phenazine derivatives.
• Employed feature selection and hyperparameter optimization for improved prediction accuracy.
• Validated models on an external test set with diverse molecular structures and functional groups.

Main Results:

• Achieved high prediction accuracies (R² > 0.74) on the external test set.
• Models successfully predicted redox potentials for derivatives with multiple functional groups, despite training on single-group molecules (R² > 0.7).
• Demonstrated unprecedented predictive performance for redox potentials using a small, simple dataset.

Conclusions:

• The hybrid DFT-ML approach effectively predicts redox potentials, reducing computational and experimental costs.
• Identified promising phenazine derivatives for green energy storage systems.
• This methodology can significantly accelerate materials discovery for RFBs.