Data mining and utilization of experimental porous materials data

Abstract

Porous materials are in the spotlight in various fields such as gas storage and separation and catalysts due to their large specific surface area and pore volume. In particular, various synthetic attempts have been made for metal-organic frameworks (MOFs) due to their high tunability. The number of MOFs reported to the Cambridge Structural Database (CSD) has increased exponentially to more than 100,000. Due to the paradigm shift of research and the exponential increase in the number of porous materials, data science research on porous materials is being actively conducted. However, various problems often arise due to the absence of organized data and data inconsistency between experiments and simulation. In this study, we propose a text-mining algorithm so that experimental data for data science can be extracted from published papers. In addition, the number of experimental data are not sufficient, so the calculated data are used together. Confirming that there is inevitably a difference between these calculated data and the experiment, differences are quantified by comparing X-ray diffraction data using earth mover’s distance (EMD), and a methodology for predicting experimental data from simulation data is presented. Finally, by predicting the adsorption isotherm and surface area through the proposed methodology, we suggest to the computational scientists that the numerical analysis of the X-ray diffraction pattern should be preceded along with pretreatment such as structural optimization.