Atomic Property Weighted Radial Distribution Functions Descriptors of Metal–Organic Frameworks for the Prediction of Gas Uptake Capacity

Abstract

Metal–organic frameworks (MOFs) are porous materials with exceptional host–guest properties with huge potential for gas separation. The combinatorial design of MOFs demands the in silico screening of the nearly infinite combinations of structural building blocks using efficient computational tools. We report here a novel atomic property weighted radial distribution function (AP-RDF) descriptor tailored for large-scale Quantitative Structure–Property Relationship (QSPR) predictions of gas adsorption of MOFs. A total of ∼58,000 hypothetical MOF structures were used to calibrate correlation models of the methane, N2, and CO2 uptake capacities from grand-canonical Monte Carlo (GCMC) simulations. The principal component analysis (PCA) transform of the AP-RDF descriptors exhibited good discrimination of MOF inorganic SBUs, geometrical properties, and more surprisingly gas uptake capacities. While the simulated uptake capacities correlated poorly to the void fraction, surface area, and pore size, the newly introduced AP-RDF scores yielded outstanding QSPR predictions for an external test set of ∼25,000 MOFs with R 2 values in the range from 0.70 to 0.82. The accuracy of the predictions decreased at low pressures, mainly for MOFs with V2O2 or Zr6O8 inorganic structural building units (SBUs) and organic SBUs with fluorine substituents. The QSPR models can serve as efficient filtering tools to detecting promising high-performing candidates at the early stage of virtual high-throughput screening of novel porous materials. The predictive models of the gas uptake capacities of MOFs are available online via our MOF informatics analysis (MOFIA) tool.

Measuring local volume fraction, long-wavelength correlations, and fractionation in a phase-separating polydisperse fluid

“…In real space, a pair of 'weighted RDFs' ξ 2 (r) and ξ 1 (r) can be used (a related idea occurs in the use of 'RDF descriptors' for chemical structure 38,39 ). The first shows up fractionation irrespective of how the phases differ, but does not show the direction of fractionation.…”

Computational development of the nanoporous materials genome

“…There have been several recent studies dedicated to answering this question using models developed by machine learning 32,45,114,[131][132][133][134] , a field that is becoming extremely powerful in materials science 135,136 . It was shown that 1-dimensional geometric descriptors are able to successfully predict adsorption at high pressures 134,137 and low temperatures 114 , using representative datasets of materials to train machine learning models. These models are however, rather poor at predicting performance at lower gas densities, in pressure regions of 0 -1 bar 134 .…”

“…It was shown that 1-dimensional geometric descriptors are able to successfully predict adsorption at high pressures 134,137 and low temperatures 114 , using representative datasets of materials to train machine learning models. These models are however, rather poor at predicting performance at lower gas densities, in pressure regions of 0 -1 bar 134 . This is likely due to the oversimplified nature of these descriptors, when a more detailed representation of the chemistry and pore shape is needed.…”

“…This is likely due to the oversimplified nature of these descriptors, when a more detailed representation of the chemistry and pore shape is needed. Indeed, the development of complex materials descriptors that include both chemical and geometric features 134 appears to improve the overall success of machine learning models. Notably, however, the model in Ref 138 .…”

“…if we are given a structure, can we predict its performance, say in CO 2 separations, based on its surface area, pore size, or pore volume? There have been several recent studies dedicated to answering this question using models developed by machine learning 32,45,114,[131][132][133][134] , a field that is becoming extremely powerful in materials science 135,136 . It was shown that 1-dimensional geometric descriptors are able to successfully predict adsorption at high pressures 134,137 and low temperatures 114 , using representative datasets of materials to train machine learning models.…”