(502g) Intensive Chemical Descriptors for Nanoporous Crystalline Materials

Molecular descriptors play a key role in the development of predictive structure-property/activity relationships (QSP/ARs) that aid molecular discovery/design. Although thousands of molecular descriptors have so far been proposed, the range of descriptors that have been explored for nanoporous crystalline materials such as metal-organic frameworks (MOFs) has been limited to few mostly macroscopic geometrical descriptors. In this study, we introduce a new robust procedure to generalize the thousands of descriptors that already known from molecular systems to crystalline nanoporous materials. We will show how these descriptors have enabled machine-learning approaches to efficiently and accurately predict adsorption isotherms for diverse collections of thousands of molecules in libraries of thousands of MOFs.