(678e) Assessing the Synthetic Feasibility of Zeolite-like Materials Using Zeonet

Zeolites are nanoporous materials widely used as catalysts and adsorbents in the chemical industry. Although hundreds of thousands of hypothetical framework structures have been predicted,¹ only ~250 zeolites have been synthesized experimentally so far. Given the nontrivial task of discovering the synthetic route for a new zeolite, it is unclear whether the large gap should be attributed to unknown synthetic recipes or to intrinsic differences between real and current generations of hypothetical zeolites. To address this question, researchers have used both physical descriptors (e.g., analysis of the distribution of geometric parameters²) and data-science-based approaches (e.g., classification using support-vector machine and the SOAP descriptors³) to distinguish real and hypothetical zeolites, with a false positive rate of 20% and 11%, respectively. We have previously developed ZeoNet, based on 3D convolutional neural networks and volumetric distance grids, and shown that it greatly outperforms 2D multi-view images, point clouds, graph neural networks in predicting the adsorption of a long-chain hydrocarbon (n-octadecane) in all-silica zeolites. Here, we assess the ability of the ZeoNet representation in distinguishing materials structures. All known zeolite structures catalogued by International Zeolite Association (IZA), including both (alumino)silicates and aluminophosphates, were digitized and used to train ZeoNet for the classification task, which achieved a false positive rate of 0.8%. We hypothesize that the misclassified hypothetical zeolites are more likely to be eventually synthesizable. These quasi-realistic structures were analyzed using previously suggested descriptors, revealing the intriguing ability of ZeoNet to capture non-obvious structural differences between real and hypothetical zeolites.

(1) Pophale, R.; A. Cheeseman, P.; W. Deem, M. A Database of New Zeolite -like Materials. Physical Chemistry Chemical Physics 2011, 13, 12407–12412.

(2) Perez, J. L. S.; Haranczyk, M.; Zimmermann, N. E. R. High-Throughput Assessment of Hypothetical Zeolite Materials for Their Synthesizeability and Industrial Deployability. Zeitschrift für Kristallographie - Crystalline Materials 2019, 234, 437–450.

(3) A. Helfrecht, B.; Pireddu, G.; Semino, R.; M. Auerbach, S.; Ceriotti, M. Ranking the Synthesizability of Hypothetical Zeolites with the Sorting Hat. Digital Discovery 2022, 1, 779–789.