Interpretable Machine Learning-Based Predictions of Methane Uptake Isotherms in Metal–Organic Frameworks
Rishi Gurnani, Zhenzi Yu, Chiho Kim, David S. Sholl, Rampi Ramprasad
Abstract
Tuning the structure of metal–organic frameworks (MOFs) is a promising pathway toward the development of high-performing materials for methane storage. To aid such discoveries, we introduce techniques for the machine-learned prediction of methane isotherms in MOFs. We demonstrate that our predictors surpass prior benchmarks. We use these models to search for novel (from both a structural and chemical point of view), high-performing MOFs and test them using density functional theory (DFT)-based structural relaxation and molecular simulation of methane adsorption. These simulations reveal that our model generalizes to chemistries not seen during training. One novel candidate, predicted to surpass the 2008 world record for volumetric methane uptake in MOFs, is proposed. Our simulations also reveal that DFT relaxation has a systematic effect on the uptake value. Finally, we interpret the models to discover and present potential MOF–methane uptake structure–property relationships.