High-Throughput Computational Screening and Machine Learning Model for Accelerated Metal–Organic Frameworks Discovery in Toluene Vapor Adsorption

Some hazardous gases, like toluene vapor, have caused serious environmental pollution. The adsorption of toluene using metal–organic frameworks (MOFs) has been considered a useful mechanism to reduce environmental pollution. High-throughput computation using the grand canonical Monte Carlo (GCMC) approach was used to screen high-performance MOFs from the CoRE MOF database. A total of 802 MOFs are selected with a toluene uptake larger than HKUST-1 (6.34 mmol/g at 1900 Pa, 298 K) and CMOF-3b ([(L3b)Cu2]n, L3b = 4,4′,4″,4‴-(2,2′-dihydroxy-1,1′-binaphthalene-4,4′,6,6′-tetrayl)), showing the highest toluene vapor adsorption capacity (25.57 mmol/g). Approximately 80% of high-performance MOFs contain open metal sites. Further analyses of the quantitative structure–property relationships reveal that the MOF adsorption capacity for toluene could be primarily correlated with gravimetric surface area and void fraction. Moreover, using the HKUST-1 as a template, center-metal element replacement is suggested to be effective in improving toluene vapor adsorption. Finally, based on our previously proposed MOF-CGCNN algorithm, a regression model is developed to predict toluene adsorption capacity. Combined with high-throughput GCMC calculation, the machine learning model is applied to screen a larger MOF database (containing 137,953 hypothetical MOFs), which accelerates the virtual discovery of new high-performance candidate MOFs for toluene adsorption. The proposed strategy will be useful in material design or discovery for reducing toluene, thereby benefitting the environment.