Machine Learning Models for Recognizing High Performing Metal Organic Frameworks for Natural Gas Purification

Metal-organic frameworks have gained interest because of their exceptionally high porosities and chemically tunable structures. In this direction, recent In-silico generation of libraries with thousands to millions of hypothetical materials and subsequently high-throughput (HT) computational screening as a part of material discovery process, illustrate the demand of more efficient computational tools compared to traditional grand canonical Monte Carlo (GCMC) simulations which can be resource-demanding. Here, we report machine learning classifier models for CO2/CH4 separation parameters that utilize separately the Voronoi hologram and Atomic Property-Weighted Radial Distribution Function (AP-RDF) descriptors. We compare their performance with previously reported classifiers models. For this, the classifiers are trained on 32,450 MOFs and then are tested on the 292,050 MOFs that are not part of the training set. From the comparison, it is found that including AP-RDF and Voronoi hologram descriptors into the classifiers improves the performance of classifiers by 20% in capturing high-performing MOFs

کلیدواژه ها:

machine learning ، nanoporous solids ، pressure swing adsorption ، QSAR ، gas separation