MOFGalaxyNet: a social network analysis for predicting guest accessibility in metal–organic frameworks utilizing graph convolutional networks

Abstract

Metal–organic frameworks (MOFs), are porous crystalline structures comprising of metal ions or clusters intricately linked with organic entities, displaying topological diversity and effortless chemical flexibility. These characteristics render them apt for multifarious applications such as adsorption, separation, sensing, and catalysis. Predominantly, the distinctive properties and prospective utility of MOFs are discerned post-manufacture or extrapolation from theoretically conceived models. For empirical researchers unfamiliar with hypothetical structure development, the meticulous crystal engineering of a high-performance MOF for a targeted application via a bottom-up approach resembles a gamble. For example, the precise pore limiting diameter (PLD), which determines the guest accessibility of any MOF cannot be easily inferred with mere knowledge of the metal ion and organic ligand. This limitation in bottom-up conceptual understanding of specific properties of the resultant MOF may contribute to the cautious industrial-scale adoption of MOFs.Consequently, in this study, we take a step towards circumventing this limitation by designing a new tool that predicts the guest accessibility—a MOF key performance indicator—of any given MOF from information on only the organic linkers and the metal ions. This new tool relies on clustering different MOFs in a galaxy-like social network, MOFGalaxyNet, combined with a Graphical Convolutional Network (GCN) to predict the guest accessibility of any new entry in the social network. The proposed network and GCN results provide a robust approach for screening MOFs for various host–guest interaction studies.

Accelerated Discovery of Metal–Organic Frameworks for CO2 Capture by Artificial Intelligence

“…ML studies on zeolites, porous carbons, MOFs, COFs, and their composites generally predicted materials’ CO 2 adsorption and separation properties based on the easily accessible structural features of materials, such as surface areas and element types. New tools were developed to predict guest accessibility of any given MOF from the chemical features, the organic linkers, and the metal ions. , Recent efforts showed that introducing new descriptors can offer time efficiency and more accurate predictions. For example, energy-based descriptors, including Gibbs free energy and Boltzmann weighted energy distributions of xenon (Xe) and krypton (Kr) gases, were demonstrated to be more important for determining Xe/Kr selectivities of MOFs compared to their structural and chemical features .…”

Multiscale Modeling of Physical Properties of Nanoporous Frameworks: Predicting Mechanical, Thermal, and Adsorption Behavior

“…Currently, the growing amount of structural and simulation data is generating a great deal of enthusiasm for creating machine learning models capable of predicting the properties of new materials. Over the past few years, a series of classes of ML models (simple ML models for regression, convolutional neural networks, graph neural networks, and transformers for more complex data) have emerged, demonstrating their growing effectiveness in predicting properties such as adsorption uptakes in MOFs, − see Figure . Authors typically concentrate on assessing their model’s performance using various metrics on a validation data set.…”