Data-Driven Methods to Accelerate the Design of Metal-Organic Frameworks

Bucior, Benjamin James

doi:https://doi.org/10.21985/n2-py89-3d83

Work

Data-Driven Methods to Accelerate the Design of Metal-Organic Frameworks

Public

Download PDF

Metal-organic frameworks (MOFs) are a class of nanoporous materials with highly tunable pore shape and chemistry. They are synthesized in a "building block" approach to form crystalline porous materials, which have been explored for diverse applications including gas storage and separations. Given the enormous size of the MOF design space, new methods are needed to rapidly match these important applications to suitable MOF compositions. This dissertation presents advances in leveraging large-scale MOF data to facilitate rapid design and understanding trends. First, a fast, predictive model was developed to correlate hydrogen uptake with the MOF-hydrogen energy landscape. The model is accurate, robust, interpretable, and three orders of magnitude faster than detailed molecular simulations, enabling us to quickly select one MOF for experimental synthesis out of 55,000 candidates. Second, two new cheminformatics formats, named MOFid and MOFkey, were designed to systematically identify MOFs and label their data. An open-source code was written with new algorithms to deconstruct MOFs into their building blocks and underlying topological nets. Developing these cheminformatics methods for MOFs made it possible to rapidly query structural databases and obtain new insights, such as structure-property relationships. Third, MOF structure data was cleaned up and analyzed as part of the curation of a computation-ready database of experimentally reported structures. A semi-automated workflow was developed to rebuild and restore disordered crystal structures that had previously been discarded from the database. The database was structurally diverse, as confirmed by a topological analysis. A MOF fingerprint based on the molecular formula, density, and other textural properties was developed to deduplicate structures, leading to the finding that approximately 80% of the structures corresponded with unique MOFs. Finally, the methods above have been extended to build a platform that can propose new MOF structures. The MOFid format was modified to a compact scheme more amenable to machine learning and reconstructing crystal structures. Exploration of the MOF design space provided insight on common MOF families and illuminated opportunities for future design. Overall, combining the strengths of molecular simulation, data science, and experiment make it possible to accelerate the design process and efficiently unlock the potential of metal-organic frameworks for energy applications.

Creator