Inverse Design of Metal‐Organic Frameworks for CH ₄ /N ₂ Separation Enabled by Coupled Machine Learning and Genetic Algorithms

Metal-organic frameworks (MOFs) have emerged as promising candidates for gas separation, yet the vastness of their structural design space renders experimental screening prohibitively time- and resource-intensive. Recent advances in machine learning (ML) technology offer powerful alternatives for accelerating MOF discovery through data-driven prediction. In this work, a high-accuracy ML model with a Tangent Adaptive Genetic Algorithm (TAGA) is integrated to enable inverse design of MOFs for CH₄/N₂ separation. The ML model, trained on structural features including topology, metal/organic secondary building units, and functional groups, is embedded within the TAGA framework to efficiently navigate the high-dimensional chemical space. Analysis of the evolutionary trajectory reveals that MOFs featuring the fsc topology and ligands such as pyrene, anthracene, and naphthalene consistently exhibit superior CH₄/N₂ selectivity. Based on these high-performance genotypes, a series of MOF structures are constructed, among which the top-performing candidate achieves an IAST selectivity of 15.92 and a CH₄ uptake of 2.47 mmol g^-1. This study highlights a paradigm shift from trial-and-error screening toward goal-directed materials design, offering a generalizable pathway for developing next-generation separation materials.