Predicting Partial\nAtomic Charges in Metal–Organic\nFrameworks: An Extension to Ionic MOFs

Molecular\nsimulation is an invaluable tool to predict and understand\nthe usage of metal–organic frameworks (MOFs) for gas storage\nand separation applications. Accurate partial atomic charges, commonly\nobtained from density functional theory (DFT) calculations, are often\nrequired to model the electrostatic interactions between the MOF and\nadsorbates, especially when the adsorbates have dipole or quadrupole\nmoments, such as water and CO<sub>2</sub>. Machine learning (ML) models\nhave been previously employed to predict partial charges and avoid\nthe computational cost associated with DFT calculations. However,\nprevious ML models suffer from small training data sets, which limit\ntheir scope of application. In this work, we introduce two novel machine\nlearning models, PACMOF2-neutral and PACMOF2-ionic, aimed at predicting\nthe density-derived electrostatic and chemical (DDEC6) partial atomic\ncharges for both neutral and ionic MOFs. These models not only yield\nDFT-level accuracy at a fraction of the computational cost but also\ndemonstrate a remarkable improvement in prediction of adsorption,\nas validated with grand canonical Monte Carlo simulations. The robustness\nand fast computational time of the PACMOF2 models, along with their\ntransferability to other porous materials such as covalent organic\nframeworks and zeolites, underscores their potential in high-throughput\nscreening of MOFs for diverse applications.