Machine Learning-Guided Pore Engineering of Metal-Organic Frameworks for Ultrahigh Volumetric Methane Storage

The high volumetric storage of methane under mild conditions remains a central challenge for the practical implementation of adsorbed natural gas technologies. Here, we report a machine learning (ML)-guided strategy that integrates large-scale computational screening with experimental pore engineering to identify and optimize metal-organic frameworks (MOFs) for high volumetric methane working capacity. An Extra Trees Regressor model was trained using adsorption data obtained through grand canonical Monte Carlo simulations on the structures from the CoRE MOF 2024 database, achieving excellent predictive accuracy (R² = 0.96). ML screening highlighted UMCM-4, MUF-8, and MOF-5 as top-performing and experimentally accessible candidates, whose measured volumetric methane working capacities at 298 K and 5-100 bar (225, 225, and 231 cm³_(stp) cm^-3) closely match ML predictions. Feature-importance analysis reveals that porosity, pore volume, and framework density govern volumetric methane storage through a nonmonotonic interplay, defining an optimal range of 0.78-0.86 for porosity, 1.20-2.00 cm³ g^-1 for pore volume, and 0.40-0.70 g cm^-3 for framework density. Guided by these insights, linker functionalization was employed to finely tune structural parameters, leading to ultrahigh volumetric methane working capacities of 237 cm³_(stp) cm^-3 for MUF-8-CH₃ and MUF-8-C₄H₄, surpassing all previously reported MOFs. This ML-guided pore engineering paradigm provides a generalizable route for the rational design of high-performance methane adsorbents.