Advancements in Machine Learning Predicting Activation and Gibbs Free Energies in Chemical Reactions

Abstract

Machine learning has revolutionized computational chemistry by improving the accuracy of predicting thermodynamic and kinetic properties like activation energies and Gibbs free energies, accelerating materials discovery and optimizing reaction conditions in both academic and industrial applications. This review investigates the recent strides in applying advanced machine learning techniques, including transfer learning, for accurately predicting both activation energies and Gibbs free energies within complex chemical reactions. It thoroughly provides an extensive overview of the pivotal methods utilized in this domain, including sophisticated neural networks, Gaussian processes, and symbolic regression. Furthermore, the review prominently highlights commonly adopted machine learning frameworks, such as Chemprop, SchNet, and DeepMD, which have consistently demonstrated remarkable accuracy and exceptional efficiency in predicting both thermodynamic and kinetic properties. Moreover, it carefully explores numerous influential studies that have notably reported substantial successes, particularly focusing on predictive performance, diverse datasets, and innovative model architectures that have profoundly contributed to enhancing computational chemistry methodologies. Ultimately, this review clearly underscores the transformative potential of machine learning in significantly improving the predictive power for intricate chemical systems, bearing considerable implications for both cutting-edge theoretical research and practical applications.