(689d) A Computational Roadmap for the De Novo Design of Organic Structure Directing Agents for Zeolite Synthesis | AIChE

(689d) A Computational Roadmap for the De Novo Design of Organic Structure Directing Agents for Zeolite Synthesis


Xie, M. - Presenter, Imperial College London
Hoffman, A., University of Florida
Schwalbe-Koda, D., Massachusetts Institute of Technology
Reyes, O. S., Massachusetts Institute of Technology
Paris, C., Instituto de Tecnologia Quimica
Moliner, M., ITQ (CSIC-UPV)
Gomez-Bombarelli, R., Massachusetts Institute of Technology
Zeolites are nanoporous materials that are heavily used in industrial catalysis and separations. Despite their vast structural and chemical diversity, limited structures and compositions are accessible due to strict synthesis windows that require substantial experimental effort to determine. A key driver of selectivity in zeolite synthesis is organic structure directing agents (OSDAs). OSDAs template specific crystalline structures (frameworks) during crystallization, but are challenging to design a priori and often costly to synthesize. In this work, we illustrate a fully computational work flow for the design of de novo OSDAs. A machine learning model is trained on over half a million binding affinities between molecules and zeolites and used to screen over a million different quaternary and di-quaternary molecules—enumerated from commercially available building blocks—for potential zeolite-OSDA matches, an exploration orders of magnitude larger than previous efforts [1]. An automated docking pipeline, benchmarked in previous work [2], verifies the binding affinities for promising molecules. Finally, shape and energy metrics are used to curate a final pool of synthesizable OSDA candidates for a targeted framework. We demonstrate the capability of this computational workflow through two syntheses of cage-based zeolites with previously inaccessible Si/Al ratios using novel OSDAs. We also highlight some insights into structural similarities between frameworks from a molecular perspective using the predicted binding energies. This work demonstrates the capability of using surrogate models and intelligent screening to vastly expand the space for realizable, novel material design in a cost-efficient manner.


[1] Proc. Natl. Acad. Sci., 2019, 116, 3413–3418

[2] Science, 2021, 374, 308–315