Machine Learning-Guided Prediction of Activation Energies for Catalyst Design in the Water Gas Shift Reaction

The water gas shift (WGS) reaction is a critical process in various industrial applications, with its rate often governed by the activation energy of water dissociation on metal-based catalysts. In this work, we develop a machine learning model to predict the activation energy for water dissociation across diverse catalytic surfaces, including monometallic, bimetallic, and trimetallic surfaces, metallic nanotubes, and platinum nanoparticles. The model links activation energy to easily computable descriptors, such as adsorption energies or distances of the involved reaction species, derived from density functional theory (DFT) calculations. Our multilayer perceptron-based regression model demonstrates exceptional predictive accuracy. A Shapley additive explanation (SHAP) analysis discloses that all input variables contribute positively to the activation energy, with the adsorption energy of reaction products identified as the most influential feature. Remarkably, our model enables the prediction of activation energy for water dissociation and WGS on metal-based catalysts not included in the training set by simply DFT computing adsorption energies and distances, significantly reducing computational effort. This approach thus provides a powerful tool for the rapid screening and optimization of catalysts for the WGS reaction, paving the way for more efficient industrial processes.