Science https://doi.org/10.1126/science.abh3350 (2021)

Zeolites are microporous materials, sometimes referred to as molecular sieves, that often serve as adsorbents and catalysts. Zeolite versatility is largely attributed to their topological diversity, which is the result of phase competition between metastable polymorphs (that is, materials that take on more than one form or crystal structure). In order to manage the zeolite topology, organic structure-directing agents (OSDAs) can be used to guide the formation of particular pore or channel types during the synthesis of a zeolite. Having this control can be particularly important for tuning the zeolite topology to be more robust toward a particular application, such as catalysis.

Previous design algorithms can lead to OSDAs that fail to crystalize the desired zeolite structure, or that generate molecules with low synthetic accessibility. The use of literature analysis, that is, evaluating experiments that have been reported in the literature with varying success, can assist in avoiding expensive and complex zeolite synthesis. However, to date, few datasets containing OSDA–zeolite affinities are available, and manual literature searches have been limited to tens of papers, which restricts the ability to explain phase selectivity and synthesis prediction. Manuel Moliner, Rafael Gómez-Bombarelli and colleagues address this problem by employing high-throughput atomistic simulations with literature mining and human–computer interactions.

Initially, automated extraction tools were used to obtain information about OSDAs from literature, which resulted in over half a million zeolite–OSDA pair predictions from varying frameworks and OSDAs. From this, phase selectivity was quantified by a new metric called the templating energy, which takes into account the competitivity (how close a framework is to the best host for a given OSDA) and directivity (how close a molecule is to the best OSDA for a given framework) of a framework. The templating energy was shown to outperform the binding energy, which is the typical metric for assessing OSDA–zeolite affinity, in selecting molecules for a given zeolite for about 70% of the structures across over 1,000 papers. Even so, the templating energy did not perform well in predicting the best hosts for a given OSDA. To further address this limitation, a database of zeolite–OSDA pairs was generated, which allows for other physical descriptors such as OSDA shape and size to be evaluated through an interactive user interface and also enables the incorporation of chemical expertise via human–computer interaction. In three examples, shape and size metrics were used to design successful OSDAs, which were shown to suppress competition and achieve purer phases with more accessible synthesis. Finally, the study was applied to a single OSDA that balanced phase competition through shape and binding metrics to facilitate intergrowth, which can be beneficial for providing unique crystallographic environments to incorporate metallic active sites.