Fast Screening of Large Databases for Top Performing Nanomaterials Using a Self-Consistent, Machine Learning Based Approach
George S. Fanourgakis, Konstantinos Gkagkas, Emmanuel Tylianakis, George E. Froudakis
Abstract
Toward the fast and accurate identification of top performing candidates from the huge pool of nanoporous materials that can potentially be synthesized, we present and evaluate a methodology that combines, in a self-consistent way, results of molecular simulations and machine learning approaches. The aim is to reduce as much as possible the number of candidates for which time-consuming and expensive theoretical and/or experimental studies are required. As a case study, we have used the adsorption of methane at different pressures in a large database of covalent-organic frameworks and in a small, but highly structural and chemical diverse database of experimentally synthesized metal–organic frameworks. It is found that when the self-consistent approach is combined with a machine learning algorithm, for which accurate descriptors are used, most of the top performing materials are identified, requiring only a small amount of reference data. The sensitivity of the algorithm to its underlined parameters is investigated and strategies for optimal choices are suggested.