An exhaustive mapping of zeolite-template chemical space

Zeolites are industrial catalysts and adsorbents whose synthesis usually employs specific molecules known as organic structure-directing agents (OSDAs). The OSDA’s templating effect is pivotal in determining the zeolite polymorph formed and its physicochemical properties. However, de novo design of selective OSDAs is challenging because of the diversity and size of the zeolite-OSDA chemical space. Here, a computational workflow powered by machine learning enables an exhaustive exploration of the OSDA space for known zeolites. Models were developed to predict molecule-zeolite binding energies and trained on hundreds of thousands of datapoints, the largest ever library of synthetically accessible, hypothetical OSDA-like molecules was enumerated from commercially available precursors, and nearly 500 million zeolite-molecule pairs were screened. From these, two new OSDAs were identified and validated experimentally to template zeolites with unique compositions. The nearly exhaustive scale of the OSDA library and open-access data are expected to accelerate OSDA design for the entire field.