Accelerating Discovery of Water Stable Metal−Organic Frameworks by Machine Learning

Abstract

Metal-organic frameworks (MOFs) provide an extensive design landscape for nanoporous materials that drive innovation across energy and environmental fields. However, their practical applications are often hindered by water stability challenges. In this study, a machine learning (ML) approach is proposed to accelerate the discovery of water stable MOFs and validated through experimental test. First, the largest database currently available that contains water stability information of 1133 synthesized MOFs is constructed and categorized according to experimental stability. Then, structural and chemical descriptors are applied at various fragmental levels to develop ML classifiers for predicting the water stability of MOFs. The ML classifiers achieve high prediction accuracy and excellent transferability on out-of-sample validation. Next, two MOFs are experimentally synthesized with their water stability tested to validate ML predictions. Finally, the ML classifiers are applied to discover water stable MOFs in the ab initio REPEAT charge MOF (ARC-MOF) database. Among ≈280 000 candidates, ≈130 000 (47%) MOFs are predicted to be water stable; furthermore, through multi-stability analysis, 461 (0.16%) MOFs are identified as not only water stable but also thermal and activation stable. The ML approach is anticipated to serve as a prerequisite filtering tool to streamline the exploration of water stable MOFs for important practical applications.