Machine Learning Approach to Vertical Energy Gap in Redox Processes.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Biophysics

Background:

• Calculating redox properties like free energy change (ΔG) and reorganization energy relies on linear response approximation (LRA).
• Accurate predictions are hindered by challenges in conformational and vertical energy-gap sampling.
• Current methods often use computationally expensive hybrid quantum mechanical/molecular mechanical (QM/MM) calculations for energy gap sampling.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting vertical energy gaps (VEGs).
• To reduce the computational expense associated with QM/MM calculations in redox property predictions.
• To assess the performance of various ML models using features from different quantum mechanical methods.

Main Methods:

• Implemented and tested multiple machine learning models, including linear regression and extra trees regressor.
• Utilized features extracted from semiempirical and quantum mechanical (QM) methods to train ML models.
• Validated ML model performance by comparing predicted VEGs against established calculations.

Main Results:

• Simple ML models, such as linear regression, demonstrated excellent performance with a mean absolute error of approximately 0.1 eV.
• The extra trees regressor model achieved a mean absolute error of around 0.1 eV, even when using features from the most computationally inexpensive QM method.
• ML models showed high accuracy in predicting VEGs across various test systems.

Conclusions:

• Machine learning offers a computationally efficient alternative for predicting vertical energy gaps in redox processes.
• The proposed ML approach can significantly reduce the cost of calculating redox properties.
• This method holds promise for generalization to larger and more complex molecular systems, including macromolecules with intricate redox centers.