Molecular modeling-based machine learning for accurate prediction of gas diffusivity and permeability in metal–organic frameworks

Abstract

Gas diffusion determines the performance of metal-organic frameworks (MOFs) in various practical applications, including membrane-based separations, yet its experimental measurement is challenging. We presented an efficient computational framework that integrates high-fidelity molecular dynamics (MD) simulations with machine learning (ML) to predict the diffusivities of CO2, N-2, O-2, CH4, and H-2 in >18,000 synthesized and hypothetical MOFs. ML models trained on MD data accurately predicted gas diffusivities of any given MOF within minutes using only easily accessible structural and guest-related properties. We provided an interactive, user-friendly web interface for predicting diffusivities of MOFs to facilitate material selection. Leveraging ML-predicted diffusivities, we evaluated membrane-based gas separation performances of all MOFs for seven industrially important separations: CO2/N-2, CO2/CH4, N-2/CH4, H-2/CO2, H-2/CH4, H-2/N-2, and O-2/N-2. The best MOF membranes offering high selectivity and permeability were identified and analyzed by using molecular fingerprinting to reveal the critical chemical properties for designing next-generation MOFs.