Data-Driven Predictionof Structures of Metal–OrganicFrameworks

Crystal structure prediction (CSP) has proven to be an effective route for the discovery of new materials. Nonetheless, the <i>ab initio</i> techniques employed for the CSP of metal–organic frameworks (MOFs) cannot be scaled to a high-throughput mode. Here, we propose a data-driven method for addressing the current needs of computational MOF discovery. Specifically, coarse-grained neural networks were implemented to predict the underlying net topology. The models showed satisfactory performance, which was next enhanced via the limitation of the applicability domain.