High-throughput and machine learning approaches for the discovery of metal organic frameworks

Abstract

Metal-organic frameworks (MOFs) are promising nanoporous materials with diverse applications. Traditional material discovery based on intensive manual experiments has certain limitations on efficiency and effectiveness when faced with nearly infinite material space. The current situation offers an opportunity for high-throughput (HT) and machine learning (ML) approaches, including computational and experimental methods, as they have greatly improved the efficiency of MOF screening and discovery and have the capacity to deal with the enormous growth of data. In this review, we discuss the research progress in HT computation and experiments and their effect on MOF screening and discovery. We also highlight how ML-based approaches and the integration of HT methods with ML algorithms accelerate MOF design. In addition, we provide our insights on the future capability of data-driven techniques for MOF discovery, despite facing some knowledge gaps as an obstacle.

Professor Sir Anthony K. Cheetham: A half-century of transformative materials science

“…Mu et al 17 adopt a fragment charge difference method to investigate the charge transfer properties of tetrathiafulvalene-based crystals and demonstrate the significant influence of both structure and chemistry on their charge transfer properties, which provides an avenue to quantify charge transfer properties for organic crystals. Zhang et al 31 review the research progress on the discovery of MOFs through high-throughput and machine learning approaches and provide important insights into the future capability of data-driven techniques for MOF discovery.…”

Self-Driving Laboratories for Chemistry and Materials Science

“…Considering the space of possible materials and structures is intractably large, traditional approaches based on manual synthesis are insufficient to explore these materials efficiently. A number of high-throughput methods, both computational, 370,[483][484][485][486] and experimental, [487][488][489] have been developed and successfully applied to combinatorially exploring the material space. These have resulted in large datasets of possible structures and materials, as well as their measured or predicted properties, paving the way for data-driven strategies.…”

Combining computational screening and machine learning to explore MOFs and COFs for methane purification

“…Computational studies have recently played a significant role in efficiently identifying the most promising MOF materials among thousands of candidates to direct the experimental efforts to these useful structures. [31][32][33] High-throughput computational screening (HTCS) studies generally use Grand Canonical Monte Carlo (GCMC) simulations to mimic the experimental gas adsorption behavior of porous materials and Molecular Dynamics (MD) simulations to compute the diffusivity of gas molecules in the pores. These simulations provide molecular insights into the adsorption and membrane-based gas separation mechanisms of MOFs and help to establish structure-performance relations to guide the design of highperforming adsorbents and membranes.…”