Prediction of Metal–Organic Frameworks with Phase Transition via Machine Learning

Abstract

Metal−organic frameworks (MOFs) possess a virtually unlimited number of potential structures. Although the latter enables an efficient route to control the structurerelated functional properties of MOFs, it still complicates the prediction and searching for an optimal structure for specific application. Next to prediction of the MOFs for gas sorption/ separation and catalysis via machine learning (ML), we report on ML to find MOFs demonstrating a phase transition (PT). On the basis of an available QMOF database (7463 frameworks), we create and train the autoencoder followed by training the classifier of MOFs from a unique database with experimentally confirmed PT. This makes it possible to identify MOFs with a high potential for PT and evaluate the most likely stimulus for it (guest molecules or temperature/pressure). The formed list of available MOFs for PT allows us to discuss their structural features and opens an opportunity to search for phase change MOFs for diverse physical/chemical application.