(6w) A Materials Genome Approach to Metal-Organic Frameworks: From Databases to Advanced Search Algorithms

My current research activity is at the intersection between molecular modeling and data science. In this poster, I will describe my postdoctoral work on metal-organic frameworks (MOFs) in Prof. Randall Snurr’s group at Northwestern University and how these activities align with the emerging research area of data science and high-throughput molecular modeling.

In collaboration with researchers from Georgia Tech and Lawrence Berkeley National Laboratory, I have created a database of thousands of computation-ready, experimentally-derived metal-organic frameworks. Construction of the database required removal of impurities and solvents using graph-labeling and cluster-detection algorithms and text-mining. The atomic coordinates of the thousands of structures will soon be made available to the public through the Nanoporous Materials Genome Center. The database is expected to grow as new structures become available in the literature. Following the compilation of thousands of MOFs, we have computationally screened these materials using atomistic grand canonical Monte Carlo simulations to find the top performing MOFs for application in methane storage and hexane isomer separation as an initial demonstration of the utility of the database.

An emerging area in computational chemistry is the high-throughput computational screening of large numbers of structures to identify promising structures and to extract useful structure-property relationship for a given application. MOFs have been a particularly good target for such an approach because of the large number of synthesized structures in the literature. The early high-throughput computational screening of MOFs has been carried out in a brute-force manner, where a number of performance evaluations (e.g., adsorption isotherm at a given pressure) are carried out for every structure in the database. What enables such a high-throughput computational screening approach are several key simplifying assumptions, such as rigid frameworks and the use of generic force field parameters and classical atomistic simulation methods. The new frontiers of high-throughput molecular modeling will involve the removal of such simplifying assumptions and usage of high-level computational chemistry methods (e.g., DFT and QM/MM). These advanced approaches will require large amounts of computational resources, making it impossible to apply them for thousands of structures. 

One way to avoid expensive performance evaluations for thousands of structures is to employ advanced search algorithms for metal-organic frameworks. I have developed an evolutionary search algorithm and applied it to a database of hypothetical MOFs to effectively find top MOFs in terms of importance performance criteria such as methane delivery capacity and volumetric surface area. The algorithm that has been developed here will not only be useful in terms of locating the best existing structures in a database but could help to find a new, as-yet synthesized MOFs for gas storage and separation application in the future. The approach could also be extended to optimize search for promising structures in other class of materials, such as metallic glasses that finds a widespread use in electronics and biomedical applications.

My future research direction will focus on the emerging area of high-throughput computation of metal-organic frameworks and other class of materials, with an emphasis on development of a platform for data science approaches and application of molecular modeling to search for best performing materials that could be used in energy and pharmaceutical applications (e.g., gas storage and drug delivery).