Machine Learning-Assisted Exploration of Chemical Space of MOF-5 Analogs for Enhanced C2H6/C2H4 Separation

Abstract

Adsorptive separation using C₂H₆-selective adsorbents can produce high-purity C₂H₄ directly, making it an energy-efficient separation method with the potential to replace cryogenic distillation. Although many C₂H₆-selective MOFs have been reported, developing MOFs with both large C₂H₆ adsorption capacity and high C₂H₆/C₂H₄ selectivity remains challenging. Herein, we present a machine learning–assisted molecular simulation strategy to explore the C₂H₆/C₂H₄ separation capability of pcu-MOFs isoreticular to MOF-5. The eXtreme gradient boosting (XGBoost) algorithm showed high accuracy in predicting the C₂H₆/C₂H₄ selectivity and C₂H₆ uptake, where Henry coefficient ratio (S⁰) and Henry coefficient of C₂H₆ (K(C₂H₆)) were identified as key factors. We further synthesized the top-performing MOF termed A-66 and experimentally verified its large C₂H₆ adsorption capacity and excellent C₂H₆/C₂H₄ separation performance. This work provides a valuable strategy for exploring the chemical space of MOF-5 analogs and identifying promising candidates for the efficient purification of C₂H₄ from C₂H₆/C₂H₄ mixtures.