Can We Really Discover New Nanoporous Materials on the Computer?

Metal-organic frameworks (MOFs) are porous materials that can be applied to modern technologies in energy storage, gas separation, carbon dioxide capture, catalysis, and many other areas.

There are an almost unlimited number of MOFs that could be synthesized from different combinations of molecular building blocks, and the Cambridge Structural Database contains around 100,000 MOF (or MOF-like) materials that have already been synthesized, creating space for many more applications. That high number, however, also comes with a catch: it can be difficult to rapidly identify the most promising materials, among millions of possibilities, for a particular application. 

In a review article published January 9 in Nature Energy, Northwestern Engineering’s Randall Snurr has surveyed the role of predictive computational modeling for the discovery of MOF adsorbents for separation and storage of energy-relevant molecules such as hydrogen and carbon dioxide. Snurr, John G. Searle Professor of Chemical and Biological Engineering, shared his insights in “Progress toward the Computational Discovery of New Metal-Organic Framework Adsorbents for Energy Applications.” His co-authors – both former post-docs from Snurr’s group – are Peyman Z. Moghadam from University College London and Yongchul G. Chung from Pusan National University. 

In the paper, the authors make clear that efficient identification of top-performing MOFs is a key to increasing their practical utility.

For some applications, Snurr writes, top-performing MOF candidates can be found via computer simulations that predict the adsorption of molecules such as hydrogen, methane, and carbon dioxide. A widely used simulation code for performing such simulations was co-developed by Snurr and his research group.

“This can be quite useful for quickly assessing the performance of existing materials for new applications,” Snurr wrote. “With the available databases of existing MOF structures, these simulations can be performed in a high-throughput mode to quickly find appropriate materials for a desired application.”

The 100,000 or so MOFs in the Cambridge database are just a fraction of the trillions of MOFs that could be synthesized. This has led researchers to create hypothetical (or proposed) MOFs on the computer and then simulate their properties to suggest new structures to be synthesized in the lab. Snurr’s group has developed key algorithms and codes for generating these hypothetical MOFs. The article discusses several examples where simulations have taken place prior to synthesis and testing of new MOFs where the results from simulations and experiments line up.

One bottleneck is that the simulations can be quite time-consuming. Computational screening, Snurr writes, has been used to screen up to 500,000 materials, a number that’s feasible for adsorption of smaller molecules such as carbon dioxide or hydrogen. Unfortunately, that approach is not workable with time-consuming simulations or ones that require quantum mechanical calculations.

“Given the large number of possible MOF materials, more efficient ways to explore this design space are needed,” Snurr wrote.

Snurr argues that genetic algorithms or other optimization methods are promising alternatives to doing “brute force” testing of large databases. Machine learning (ML) is another technology that is starting to play a big role in sorting through which MOFs are best suited for an application, and there is still room for improvement. One area that needs improvement is feature development, Snurr argues. Another issue is that proper documentation in research papers is often missing, and the article discusses best practices for complete documentation including “the full set of materials used in the training and testing, the details of the final ML model including the hyperparameters and how they were tuned, as well as full information about the underlying molecular simulations, including force field parameters, and length of simulation.”

There are a growing number of examples where computational screening has identified new applications of existing MOFs, but finding brand new MOFs is more complicated, because MOFs can be proposed that would be difficult to synthesize in the lab. 

“Thus, the number of truly new MOFs discovered on the computer (versus finding new applications of existing materials) is still quite limited,” Snurr wrote. “Here, greater feedback and collaboration between computational researchers and synthetic MOF chemists should help the field advance more quickly in the coming years.”