Topological data analysis enhanced prediction of hydrogen storage in metal-organic frameworks (MOFs)

Abstract

Metal-organic frameworks (MOFs) have the capacity to serve as gas capturing, sensing, and storing systems. It is usual practice to select the MOF from a vast database with the best adsorption property in order to do an adsorption calculation. The costs of computing thermodynamic values are sometimes a limiting factor in high-throughput computational research, inhibiting the development of MOFs for separations and storage applications. In recent years, machine learning has emerged as a promising substitute for traditional methods like experiments and simulations when trying to foretell material properties. The most difficult part of this process is choosing characteristics that produce interpretable representations of materials that may be used for a variety of prediction tasks. We investigate a feature-based representation of materials using tools from topological data analysis. In order to describe the geometry of MOFs with greater accuracy, we use persistent homology. We show our method by forecasting the hydrogen storage capacity of MOFs during a temperature and pressure swing from 100 bar/77 K to 5 bar/160 K, using the synthetically compiled CoRE MOF-2019 database of 4029 MOFs. Our topological descriptor is used in conjunction with more conventional structural features, and their usefulness to prediction tasks is explored. In addition to demonstrating significant progress over the baseline, our findings draw attention to the fact that topological features capture information that is supplementary to the structural features.