Interaction of magnesium ion and acetate anion in bulk water: Toward high-level machine learning potential

Metal ions play crucial roles in various biological functions, including maintaining homeostasis, regulating muscle contraction, and facilitating enzyme catalysis. However, accurately simulating the interaction between metal ions and amino acid side chain analogs using high-level wave function theories remains challenging due to the significant computational costs involved. In this study, deep potential molecular dynamics (DeePMD) simulation was employed to investigate the solvation structure of the Mg²⁺-Ac⁻ ion pair in aqueous solution. To address the computational bottleneck associated with expensive quantum mechanics (QM) methods, the Deep Kohn-Sham (DeePKS) approach was utilized, which allows us to generate highly accurate self-consistent energy functionals while significantly reducing computational costs. The root mean square error and mean absolute error of energies and atomic forces indicate close agreement between DeePKS predictions and QM strongly constrained and appropriately normed (SCAN) calculations. Moreover, the neural network potential (NNP) generated using the SCAN-level dataset predicted by DeePKS exhibits higher accuracy compared to previous work, which employed a moderate BLYP functional. The potential of mean force for the Mg2+-Ac⁻ system was further examined, revealing a preference for monodentate coordination of Mg²⁺ with a ∼5.8 kcal/mol energy barrier between bidentate and monodentate geometries. Overall, this work provides a comprehensive, precise, and reliable methodology for investigating metal ions’ properties in aqueous solutions.