RNA Dynamics and Interactions Revealed through Atomistic Simulations

RNA function is deeply intertwined with its conformational dynamics. In this review, we survey recent advances in the use of atomistic molecular dynamics simulations to characterize RNA dynamics in diverse contexts, including isolated molecules and complexes with ions, small molecules, or proteins. We highlight how enhanced sampling techniques and integrative approaches can improve both the precision and accuracy of the resulting structural ensembles. Finally, we examine the emerging role of artificial intelligence in accelerating progress in RNA modeling and simulation.

💡 Research Summary

This review provides a comprehensive overview of recent advances in atomistic molecular dynamics (MD) simulations applied to RNA dynamics, covering isolated RNAs, ion‑RNA complexes, small‑molecule ligands, and RNA‑protein assemblies. The authors begin by emphasizing that RNA function depends not only on primary sequence but also on secondary and tertiary structures, whose conformational flexibility—referred to as “dynamics”—is essential for binding and catalysis. They then map the landscape of simulation methods onto the relevant time‑ and length‑scales: quantum‑mechanical (QM) calculations for chemical reactions, QM/MM hybrids for localized electronic effects, classical atomistic force fields for micro‑ to millisecond dynamics, and coarse‑grained (CG) models for large‑scale conformational transitions. Figure 1 illustrates how each method addresses distinct physical phenomena.

A central theme is the trade‑off between precision (reproducibility of results given limited sampling) and accuracy (agreement with experimental observables). The review explains that precision can be improved by longer trajectories or enhanced‑sampling techniques (metadynamics, temperature‑accelerated MD, replica‑exchange, etc.), while accuracy requires integration of experimental data. Three integration strategies are discussed: (1) ensemble refinement, where MD‑generated conformational ensembles are re‑weighted to match NMR, SAXS, or FRET data; (2) on‑the‑fly biasing, which applies experimental restraints during simulation; and (3) direct force‑field reparameterization against experimental benchmarks.

The authors evaluate the most widely used RNA force fields. AMBER’s χOL3 and CHARMM36 are highlighted as the current standards, with recent refinements that adjust hydrogen‑bond strengths, Lennard‑Jones parameters, and charge‑derivation schemes. Polarizable force fields are acknowledged for their theoretical advantages but noted as computationally demanding and still under development. Water models (TIP3P, TIP4P‑D, OPC, etc.) are also compared, emphasizing that the choice of solvent model can influence RNA stability and ion binding.

Specific RNA systems are examined to illustrate methodological progress. Canonical A‑form RNA duplexes serve as benchmarks for helical parameter validation; extensive tetramer/hexamer studies provide high‑resolution NMR data that expose force‑field deficiencies in base‑stacking and flexibility. Hairpin loops, especially the GNRA, UNCG, and CUUG tetraloops, are presented as challenging test cases where enhanced sampling combined with NMR restraints successfully reproduces heterogeneous ensembles. The review also discusses RNA‑DNA hybrids, noting that existing RNA and DNA force fields are not yet compatible enough to reproduce experimental hybrid structures.

Ion and ligand interactions receive special attention. Mg²⁺ inner‑sphere binding and K⁺‑stabilized G‑quadruplexes are modeled using QM/MM or polarizable force fields to capture charge transfer and polarization effects that classical non‑polarizable models miss.

Finally, the emerging role of artificial intelligence is explored. Deep‑learning models are being used to generate latent representations of RNA conformations, to predict force‑field parameters, and to accelerate the re‑weighting of simulation ensembles. Although still at a proof‑of‑concept stage, AI promises to streamline the integration of massive experimental datasets with simulation, potentially delivering simultaneous gains in both precision and accuracy.

In conclusion, the authors argue that the future of RNA modeling lies in integrated pipelines that combine (i) continuously refined, possibly polarizable, force fields; (ii) state‑of‑the‑art enhanced‑sampling algorithms; and (iii) systematic incorporation of experimental observables, increasingly aided by AI‑driven tools. Such holistic approaches are expected to deepen our mechanistic understanding of RNA dynamics and to accelerate the design of RNA‑targeted therapeutics.