Enhanced sampling and cryo-EM data resolve magnesium binding to RNA
Magnesium ions are essential for RNA structure but difficult to model due to slow binding kinetics and experimental limitations. We present an enhanced-sampling strategy that accelerates Mg$^{2+}$ inner-shell binding by orders of magnitude, enabling quantitative exploration of ion-binding motifs in a large ribozyme. The method combines a barrier-flattening bias with Hamiltonian replica exchange to efficiently sample multiple equivalent binding sites, and builds on an approach that achieved top performance in the CASP16 blind assessment of RNA solvation structure. Using cryo-electron microscopy maps for validation, we introduce a local analysis framework that infers the population of individual binding motifs from their agreement with experimental density, enabling site-by-site validation. We find that insufficient sampling of inner-shell binding leads to significantly poorer agreement with experiment, whereas force fields predicting different inner/outer binding equilibria remain largely indistinguishable at the current experimental resolution. These results highlight the dominant role of sampling in modelling divalent ion binding and provide a general strategy for integrating simulations with experimental data in complex biomolecular systems.