A Transformer-Based Approach To Discover Water-Stable MOFs for Ethane/Ethylene Separation

To enable industrial implementation, metal-organic frameworks (MOFs) must retain structural stability under humid conditions. Given the vast chemical space, we introduce a transformer-based language model to predict the water stability of MOFs. The model shows good predictive accuracy when validated against more than 600 experimentally reported MOFs. By integrating performance and physical-property prediction models, one MOF with both high ethane/ethylene separation performance and potential water stability was identified from a hypothetical database of over 2 million MOF structures. Multiscale calculations indicated that the outstanding separation performance of this material originates from its unique pore architecture: dumbbell-shaped channels can create strong interactions at the central constrictions, along with a layered-pillar structure providing multiple adsorption sites. This study may offer an efficient strategy for discovering industrially viable MOFs and rational guidelines for designing high-performance adsorbents.