2024 AIChE Annual Meeting

Virtual Nodes for Zeolite Property Prediction

Zeolites are crystalline, nanoporous materials that are used in industrial applications in catalysis, separation, and ion exchange. However, their synthesis and properties are complex topics that require significant study. Recently, there have been many approaches to understand zeolite with machine learning. Task-relevant virtual nodes have been shown to enhance machine learning model robustness - that is, the model's ability to produce accurate predictions when faced with data or conditions outside its training set - by incorporating additional valuable information(1). In the context of zeolites, Voronoi nodes—points in three-dimensional space that locally maximize the distance from surrounding host atoms in a crystal structure—can serve as task-relevant virtual nodes for understanding zeolites. These nodes can represent zeolite cages—nanoscale void spaces within the zeolite's crystalline framework formed by interconnected rings of silicon and oxygen atoms. These cages are crucial for understanding the docking process between zeolites and organic structure-directing agents (OSDAs) in zeolite synthesis(2), as they determine the size and shape of OSDA that can be accommodated within the zeolite structure. These Voronoi nodes show potential for improving model robustness of predicting the binding energy between zeolite and OSDA for zeolite synthesis. This study aims to explore the effectiveness of Voronoi nodes in predicting other zeolite properties. We will utilize the zeo-1 dataset (3) to generate Voronoi nodes for zeolite structures and assess the robustness of the MACE (Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields) model(4) when trained with these Voronoi nodes for energy, force, and stress prediction. Our research aims to investigate whether Voronoi nodes can enhance our understanding of the structure-property relationships in zeolites, and offer new insights into their complex behavior and characteristics.

  1. Okabe R, Chotrattanapituk A, Boonkird A, et al. Virtual node graph neural network for full phonon prediction. Nat Comput Sci. 2024;4(8):522-531. doi:10.1038/s43588-024-00661-0

  2. Schwalbe-Koda D, Gómez-Bombarelli R. Supramolecular recognition in crystalline nanocavities through Monte Carlo and Voronoi Network algorithms. The Journal of Physical Chemistry C. 2021;125(5):3009-3017. doi:10.1021/acs.jpcc.0c10108

  3. Komissarov L, Verstraelen T. Zeo-1, a computational data set of zeolite structures. Sci Data. 2022;9(1):61. doi:10.1038/s41597-022-01160-5

  4. Batatia I, Kovacs DP, Simm G, Ortner C, Csányi G. MACE: Higher order equivariant message passing neural networks for fast and accurate force fields. Adv Neural Inf Process Syst. 2022;35:11423-11436.