Predicting catalytic pathways for Thiophenol decomposition on TM-doped MoS₂: a comparative machine learning study

Thiophenol (TP), a high-toxicity compound prevalent in pharmaceuticals and industrial products, necessitates efficient catalytic decomposition methods. While two-dimensional MoS₂offers a promising large surface area for catalysis, its inert basal plane and weak TP adsorption energy (1.60 eV) limit its efficacy. To address this, we designed a single-atom catalyst via transition metal (TM) doping of MoS₂. Using first-principles calculations, we demonstrate that TM doping drastically alters the local charge density, significantly enhancing adsorption and catalytic activity for TP decomposition into H₂and H₂S. Our results identify Ni-doped MoS₂as kinetically favored and Co-doped MoS₂as thermodynamically favored for the reaction. Furthermore, we evaluated four machine learning models (linear regression, K-nearest neighbors, random forest, and gradient boosting regression trees) for predicting activation barriers and reaction energies. Random forest regression emerged as the most accurate predictor. This work provides a theoretical framework for eliminating toxic organic pollutants and establishes a machine-learning-guided strategy for accelerating catalyst screening.