Machine-Learning-Assisted High-Throughput computational screening of Metal–Organic framework membranes for hydrogen separation

The high-throughput computational screening (HTCS) and machine learning (ML) were applied to evaluate the H2 separation performances of computation-ready, experimental metal–organic framework membranes (CoRE-MOFMs). For the separation of H2/X (X = CH4, N2, H2S, O2, CO2, and He), based on the results of the structure-performance relationships by univariate analysis found that the performance of the top-performing candidates far exceeded Robeson’s upper bound. To evaluate the H2 permeability (PH2) and permselectivity, a new trade-off variable, the trade-off multiple selectivity and permeability (TMSP), was defined as the comprehensive performance of the CoRE-MOFMs. The TMSP could effectively evaluate the separation performance of CoRE-MOFMs for a variety of H2/X pairs at inifinite dilution. Eight ML methods based on supervised learning algorithms were employed to predict PH2 and TMSP, and the Gaussian process regression and random forest methods exhibited the first and second highest predictive abilities, respectively. Calculating the feature importance of each CoRE-MOFM to investigate the relationship between the number of features and the accuracy of the ML model. And a decision tree model was established following the principle of the minimum Gini coefficient, which would be accurately classified the CoRE-MOFMs based on the membrane structural descriptors; the optimal structural descriptors of MOF and the physical properties of the gas to be separated are used to establish a multivariate regression model which can predict the value of the optimal structural descriptors of MOFMs by knowing the physical properties of component X in the other H2/X separation. The quantitative structure-performance relationships on the microscopic scale could provide a theoretical basis for the screening of top-performing CoRE-MOFMs. The HTCS and ML methods could efficiently accelerate the design and the development of top-performing CoRE-MOFMs for H2 separation.