The transformative role of machine learning in advancing MOF membranes for gas separations

Abstract

Metal-organic frameworks (MOFs) have been widely recognized for their potential as gas separation membranes thanks to their unique structural properties and high performance to selectively separate different types of gas molecules. MOF membranes offer great potential to replace conventional membrane materials in addressing environmental challenges like carbon capture. Experimental fabrication and testing of a single MOF membrane, even for a single type of gas separation, requires significant resources and time. Therefore, computational modeling of MOF membranes, more specifically high-throughput molecular simulations of MOFs, for various types of gas separations has been very useful in accelerating the discovery of novel MOF membranes. With the ever-increasing number of synthesized and hypothetical MOFs, reaching up to several million material candidates, brute-force molecular simulations are no longer sufficient to comprehensively explore the vast MOF space. Integration of machine learning (ML) approaches with molecular simulations has very recently accelerated materials discovery in the field of MOF membranes. ML has been very useful not only for predicting the key membrane properties of MOF membranes such as gas permeability and selectivity but also for uncovering the hidden structure-performance correlations. Compared to experimental methods and classical molecular simulations, ML offers similar accuracy at a fraction of the cost for the design and discovery of high-performing MOF membranes. This perspective focuses on the state-of-the-art ML applications in the field of MOF membranes, discusses the recent advances in this emerging field, and addresses current challenges and future directions.