High-Throughput Computational Study of CH3I/H2O Adsorption and Separation in MTV-MOFs with Machine-Learning Analysis

Abstract

Abstract Efficient removal of radioactive methyl iodide (CH3I) from humid nuclear off-gas is critical for nuclear safety, environmental protection, and human health. Grand canonical Monte Carlo (GCMC) simulations were used to compare single-component CH3I adsorption and competitive CH3I/H2O adsorption in 560 multivariate metal–organic frameworks (MTV-MOFs). At 423.15 K and 1 bar, pMOF_333 gave the highest simulated single-component CH3I uptake of 10.01 mmol/g, whereas pMOF_360 ranked first by the working-capacity-based uptake-selectivity trade-off metric (TSN) under the specified 1:1 binary equilibrium condition, with an uptake of 8.25 mmol/g, a working capacity of 7.83 mmol/g, and a CH3I/H2O selectivity of 263.69. Water-rich, temperature-dependent, and trace CH3I/H2O/N2 simulations examined pMOF_360 beyond the principal screening condition. Post hoc machine-learning and SHapley Additive exPlanations (SHAP) analyses, together with adsorbate-distribution, interaction-energy, and Widom analyses, characterized database-internal structure–performance associations and representative framework–adsorbate interactions. These equilibrium results identify theoretical MTV-MOF candidates for further experimental and dynamic validation.