A machine learning approach to predict C-H bond activation on dual-site doped graphene catalysts.

The catalytic performance of graphene-based dual-metal-site catalysts (DMSCs) for the nonoxidative coupling of methane (NOCM) is systematically investigated using machine learning (ML)-assisted DFT calculations. To obtain insights into the methane C-H bond activation, several ML algorithms are used to predict adsorption energies ( and ) and activation barriers (E a) at each step. Initially, DFT-evaluated energies for the first C-H activation step of NOCM are used as the target variable for training various ML models. In addition, several readily available features associated with transition-metal-doped graphene are used as input features to construct a dataset, which is then used by ML models to predict the catalytic activity of DMSCs. All models are validated by performing 100 random splits of the training and test data to avoid training bias. Model validation, performed through 100 random train-test splits, identified random forest regression (RFR) and extra tree regression (ETR) as the most reliable predictors, yielding test errors of 0.33 eV for and 0.27 eV for E a. Beyond the initial C-H activation, the study explores the C-C coupling pathway leading to C2 hydrocarbon formation. Among the DMSC configurations examined, Fe/Ni and Fe/Cu exhibited the lowest coupling barriers, demonstrating their superior catalytic activity. This study demonstrates that ML models can be effectively used for catalyst screening, providing promising descriptors from input features that can accelerate high-throughput screening and rational design of efficient catalysts for methane activation and coupling reactions.