BPS2026 – Constructing atomistic RNA ensembles by enhanced sampling MD simulation

Accurately constructing RNA conformational ensembles is essential for understanding RNA function and enabling RNA-targeted drug discovery. Current computational strategies often struggle to sample lower populated states, particularly rare but functionally important conformations. Fragment-assembly approaches such as FARFAR-NMR expand accessible space efficiently, yet they frequently underrepresent the full breadth of conformational diversity. Here, we present a systematic benchmarking study of enhanced sampling molecular dynamics (MD) simulations, including temperature replica exchange MD (T-REMD), Gaussian-accelerated MD (GaMD), replica exchange GaMD (Rex-GaMD), and replica exchange with solute tempering 2 (REST2) to generate RNA conformational libraries. We evaluate these methods across three different RNA systems, including a UUCG stem-loop, the HIV trans-activating response element (TAR), and the preQ1 riboswitch. Using an optimization framework combined with experimental NMR restraint (residual dipolar couplings) measurements, we construct RNA ensembles from each conformational library and quantitatively evaluate their accuracy by comparing them to the experimental observations. Our results demonstrate that all four enhanced sampling methods outperform traditional approaches like regular MD and FARFAR to generate a more comprehensive and accurate conformational library. T-REMD achieves the highest correlation with experimental observations, albeit at a substantial computational cost. Rex-GaMD and REST2 offer computationally efficient alternatives that maintain superior performance compared to existing approaches. Notably, standard GaMD shows reduced accuracy for highly flexible systems, such as in the preQ1 riboswitch, due to intrinsic limitations in its acceleration potential when applied to systems with extensive conformational changes. In contrast, the replica exchange variants overcome these limitations through enhanced conformational sampling capabilities, achieving consistently robust performance across all tested systems. This work establishes enhanced sampling MD simulations as a tool for generating RNA conformational libraries to construct RNA ensembles, providing structural insights for therapeutic development and mechanistic understanding of RNA dynamics in biological processes.