Mechanistic Insights into Iron–Sulfur Clusters for Direct Coal Liquefaction: A Combined First-Principles and Machine Learning Study

Direct Coal Liquefaction (DCL) is a promising route for converting abundant coal resources into liquid fuels, yet its efficiency remains strongly dependent on catalyst performance. In this work, we present an integrated computational framework combining density functional theory (DFT) calculations with machine learning (ML) to investigate iron–sulfur (FeS) cluster catalysts for DCL. DFT calculations were employed to examine hydrogen-donor dissociation and coal-derived radical hydrogenation on representative FeS clusters. The results indicate that the most favorable catalytic pathways arise from the cooperation between metallic Fe sites (Fe₂) and interfacial Fe sites adjacent to sulfur (Fe₁), while sulfur atoms mainly play an indirect structural and electronic modulation role. Based on these mechanistic insights, a database containing thermodynamic and kinetic data for 636 reactions across 50 FeS cluster models was constructed. This dataset was then used to train three ML classifiers, among which the Random Forest model showed the best performance, reaching accuracies of 80% for H-donor cleavage and 93% for radical hydrogenation on the held-out test sets. SHapley Additive exPlanations (SHAP) analysis further showed that descriptors associated with Fe active-site identity were among the most influential variables in both tasks. Overall, this work provides a mechanistically informed and interpretable computational framework for understanding FeS-catalyzed DCL chemistry and for the preliminary screening of catalyst motifs within the chemical space covered by the present FeS cluster library.