Chemistry-Encoded Convolutional Neural Networks for Predicting Gaseous Adsorption in Porous Materials

Abstract

Metal–organic frameworks (MOFs) are an emerging class of materials possessing significant potential in separation and storage applications. Identifying optimal candidates from tens of thousands of MOFs that have been reported is a challenging task. To this end, machine learning (ML) represents a promising approach to facilitate the selection of best-performing MOFs. In this study, we propose a scheme to develop chemistry-encoded convolutional neural network (CNN) models to predict gaseous adsorption properties, i.e., Henry’s constants of adsorption and adsorption selectivity, in chemically diverse MOFs. To train CNN models, the MOF structures are represented by their atomic locations coupled with associated chemical information of each framework atom including the 6–12 Lennard-Jones parameters (i.e., σ and ε) and point-charge values (i.e., q). Henry’s constants of CH4 and CO2 in approximately 10 000 MOF structures computed via molecular simulations are used for training and testing. Our developed CNN models show a superior prediction accuracy. Models for zeolites are also developed for comparative purposes. Various key aspects of the CNN models, such as data augmentation and spatial resolution, are also systematically investigated for achieving high accuracy.

X-ray Diffraction and Molecular Simulations in the Study of Metal–Organic Frameworks for Membrane Gas Separation

“…Copyright 2021 American Chemical Society. (d) was reproduced with permission from ref . Copyright 2022 American Chemical Society.…”

“…The most important step in typical feature-based machine learning schemes (e.g., those mentioned above) is the identification of significant structural features for use as model inputs; however, this process is highly susceptible to human bias. Lin et al recently used a convolutional neural network (CNN) to train machines to detect structures on which to base predictions. , In their most recent study, they represented MOF structures using a framework with a spatial distribution of element parameters (i.e., σ and ε) and point-charge values (i.e., q ) pertaining to atoms located in specific positions (Figure d) . CNN models were then used to identify MOFs with high adsorption coefficients for CH 4 or CO 2 .…”

“…Lin et al recently used a convolutional neural network (CNN) to train machines to detect structures on which to base predictions. 103,104 In their most recent study, they represented MOF structures using a framework with a spatial distribution of element parameters (i.e., σ and ε) and point-charge values (i.e., q) pertaining to atoms located in specific positions (Figure 7d). 101 CNN models were then used to identify MOFs with high adsorption coefficients for CH 4 or CO 2 .…”

Deep-Learning-Based End-to-End Predictions of CO2 Capture in Metal–Organic Frameworks

“…Recently, several studies have been conducted to apply deep learning in the field of performance prediction for nanoporous materials. Lin et al used a 3D CNN model with voxels as input to predict the adsorption properties of porous nanomaterials. , Lee et al designed an ANN model using building blocks and the topology as input to predict methane storage properties . Sun et al applied metalearning to directly encode the hydrogen loading surface of nanoporous materials as input information to predict the adsorption loading of multiple materials over a wide range of pressures and temperatures …”

A Self-Evolutionary Methodology for Reverse Design of Novel MOFs

“…Such screening studies are of high computational demand and time-consuming if using an evaluation mode of traversing every structure (called the brute-force approach). Thus, the introduction of machine learning (ML) techniques into materials investigation has become a hotspot in recent years for various applications. − To the best of our knowledge, the first such study related to MOFs was reported in 2013 by the group of Woo and his co-workers, showing that a support vector machine (SVM) model combined with simple geometric descriptors can achieve high accuracy for predicting the CH 4 storage capacity of MOFs at high pressures. At the moment, diverse chemoinformatic descriptors relevant to MOFs have been created in the ML-aided computational studies, − as comprehensively summarized in recent review articles. − These new descriptors combined with conventional ones greatly facilitate the deep extraction of complicated quantitative structure–property relationships (QSPR) of MOFs, boosting the discovery efficiency of desired materials.…”