Optimal Surrogate Models for Predicting the Elastic Moduli of Metal–Organic Frameworks via Multiscale Features

Evaluating the mechanical stability of metal–organic frameworks (MOFs) is essential for their successful application in various fields. Therefore, the objective of this study was to develop optimal machine learning (ML) models for predicting the bulk and shear moduli of MOFs. Considering the effects of global (such as porosity and topology) and local features (including metal nodes and organic linkers) on the mechanical stability of MOFs, we developed multiscale features that can incorporate both types of features. To this end, we first explored descriptors representing the global and local features of MOFs from data sets of previous studies in which elastic moduli were computed. We then assessed the performance of various combinations of these descriptors to determine the optimal multiscale features for predicting the elastic moduli. The optimal surrogate models trained using multiscale features exhibited R 2 values of 0.868 and 0.824 for bulk and shear moduli, respectively. Furthermore, the surrogate models outperformed the prior benchmarks. Finally, through model interpretation, we discovered that for similar pore sizes, metal nodes are the most dominant factor affecting the mechanical properties of MOFs. We anticipate that our approach will be a valuable tool for future research on the discovery of mechanically robust MOFs for various industrial applications.