(113c) Using Data Analytics to Find Promising Metal–Organic Frameworks for Adsorption Applications

Snurr, R. Q., Northwestern University
Bucior, B., Northwestern University
Lee, S., Pusan National University
Chung, Y. G., Pusan National University
Data analytics, including methods such as machine learning, are poised to play an important role in the development of new adsorbents, especially for crystalline materials such as zeolites and metal-organic frameworks (MOFs). Recently, we have updated the Computation-Ready, Experimental (CoRE) MOF database with additional analyses and almost 10,000 additional structures, bringing the total to nearly 15,000. We report separate versions with all solvent molecules removed (ASR) and only free solvent molecules removed (FSR). Crystal structures (only for the subset that underwent significant changes during curation), pore analytics, and physical property data are included with the publicly available CoRE MOF 2019 database. A detailed analysis of open metal sites shows that a large fraction of the CoRE MOF structures (34% for FSR and 62% for ASR) contain at least one open metal site. This knowledge can be used to screen the database for selective adsorption of small molecules and catalysis applications. For example, previously, Simon and coworkers carried out computational screening of the CoRE MOF 2014 Database for Xe/Kr separation and found SBMOF-1 (CSD REFCODE: KAXQIL) to be the most selective MOF for Xe/Kr separation. In our recent follow-up work, computational screening of the CoRE MOF 2019 databases discovered structures with similar selectivity as SBMOF-1 but with higher capacity. Based on a molecular fingerprint approach that we developed, we compared the CoRE MOF 2019 Database with a database of hypothetical MOFs. We found that there are at least 16 synthesized MOFs present in the hypothetical MOF database of Wilmer et al. In addition, we report a systematic, machine-readable identification scheme for MOFs. The format, which we call MOFid, includes the MOF topology and SMILES strings for the building blocks. We anticipate that MOFid may be of considerable value for various analyses of MOFs, and we provide a few preliminary examples.