Feature-Extended Descriptor Construction for Prediction of Consecutive Elementary Reaction Energies in Methane Oxidation

Conventional descriptor construction approaches are fundamentally limited by their narrow focus on predicting adsorption energies for structurally similar intermediates, failing to address the complexity of real catalytic systems involving multiple active sites and elementary reactions. To overcome these limitations, we develop a general feature extension approach that systematically expands descriptor functionality through two key innovations: structure extension for diverse active sites and reaction extension for consecutive elementary reactions, enabled by comparative analysis of active site characteristics. Applying this approach to methane oxidation to methanol with three consecutive elementary reactions and six possible active sites (totaling 1,026 metal-exchanged zeolite catalysts), we successfully mapped the complete reaction pathways, identified optimal bimetallic catalysts (including experimentally verified Ag and Cu bimetallic active sites), and extracted fundamental structure–activity relationships. Proof-of-concept verification supports the generalizability of the feature extension approach, establishing a paradigm shift in descriptor discovery that transcends traditional single-site/single-step limitations and offering a powerful framework for complex catalytic pathway analysis and rational catalyst design.