Analyzing acetylene adsorption of metal–organic frameworks based on machine learning

Abstract

Abstract Metal-organic frameworks (MOFs) containing open metal sites are important materials for acetylene (C2H2) adsorption. However, it is inefficient or even impossible to search suitable MOFs by molecular simulation method in nearly infinite MOFs space. Therefore, machine learning (ML) methods are adopted in the material screening and prediction of high-performance MOFs. In this paper, architecture, chemical and structural features are used to analyze the C2H2 adsorption performance of the MOFs. Different ML algorithms are applied to perform classification and regression analysis to the factors affecting material adsorption. By decision tree (DT) algorithm, it is found that only PV, GSA, and Cu-OMS are sufficient to determine the high adsorption of the MOFs. Furthermore, the influence of topology on the performance of MOFs is obtained. Gradient Boosting Decision Tree (GBDT), Support Vector Machine (SVM), and Back Propagation Neural Network (BPNN), are introduced to analyze the quantitative structure-property relationship (QSPR) between C2H2 adsorption and the features of MOFs. The prediction of the GBDT model is found to have the highest accuracy, with R2 as 0.93 and RMSE as 11.58. In addition, the GBDT model is used for feature analysis, and the contribution of each feature to the performance is obtained, which is of great significance for the design and analysis of MOFs. The successful application of ML to MOFs screening greatly reduce the calculation time and provides important reference for the design and synthesis of new MOFs.