High-Throughput Screening of Metal–Organic Frameworks for Hydrogen Cyanide Capture via Molecular Simulation and Machine Learning

Hydrogen cyanide (HCN) is a highly toxic and volatile chemical hazard, for which effective and readily deployable adsorption-based mitigation materials are urgently needed. Herein, we develop an integrated high-throughput screening framework that combines Grand Canonical Monte Carlo (GCMC) simulations with machine learning (ML) to identify metal–organic frameworks (MOFs) with superior HCN adsorption performance at trace concentrations. A subset of 6,509 experimentally synthesized MOFs from the CoRE MOF database was evaluated under ternary N₂/O₂/HCN atmospheres at ppm levels. Structural, chemical, and energetic descriptors were extracted to train classification and regression models, achieving strong predictive accuracy (R² > 0.80). SHAP analysis reveals that metal identity, isosteric heat of adsorption, Henry’s coefficient, and pore structural features are the dominant factors governing HCN uptake. The screening identifies a group of high-performing MOFs, with rare-earth and alkali metal centers and open metal sites being particularly favorable. This work demonstrates an efficient simulation–ML paradigm for rationally identifying candidate adsorbents for toxic gas capture and provides interpretable design principles for developing next-generation HCN sorbents, while the predicted capacities should be interpreted as comparative upper-limit values under dry-condition and rigid-framework assumptions.