Integrating Molecular Simulations with Machine Learning to Discover Selective MOFs for CH₄/H₂ Separation

As the number of synthesized and hypothetical metal-organic frameworks (MOFs) continues to grow, identifying the most selective adsorbents for CH₄/H₂ separation through experimental or computational methods has become increasingly complex. This study integrates molecular simulations with machine learning (ML) to evaluate the CH₄/H₂ separation performance of 126605 distinct types of MOFs. Grand canonical Monte Carlo (GCMC) simulations were performed to produce CH₄ and H₂ adsorption data for synthesized MOFs at various pressures, which were then used to train ML models incorporating structural, chemical, and energetic features of the MOFs. These ML models were subsequently transferred to hypothetical MOFs, enabling the rapid and accurate screening of promising adsorbents for CH₄/H₂ separation. The top-performing MOFs were identified based on their CH₄/H₂ selectivities, and their key structural and chemical characteristics were analyzed. Synthesized (hypothetical) MOFs having narrow pores and pyridine-, histidine-, and imidazole-based (carboxylate-, benzoate-, and cubane-based) linkers demonstrated high selectivities up to 85 (115) at 1 bar and 298 K. Our findings highlight the potential of MOFs as superior alternatives to traditional adsorbent materials for CH₄/H₂ separation.