RNA conformational dynamics and the limitations of static structures
RNA molecules adopt multiple three-dimensional shapes to perform cellular tasks. These conformational changes are not captured by static structures obtained from methods such as X-ray crystallography. Without dynamics the temporal behavior of RNA remains partially understood.
RNADynBench a standardized RNA dynamics benchmark
RNADynBench provides a standardized collection of RNA molecular dynamics trajectories. The benchmark contains 2585 quality-controlled 100-ns all-atom trajectories with leakage-controlled splits. This addresses the scarcity of large-scale RNA dynamics data and enables unified evaluation of generation and understanding models.
RNADynNet a unified model for trajectory generation and fingerprinting
RNADynNet is a unified model that learns both trajectory generation and dynamics fingerprinting from a single conformer. It uses a shared backbone for both objectives and combines coordinate denoising single-frame-to-trajectory alignment and physical grounding.
Coordinate denoising single-frame-to-trajectory alignment and physical grounding
Coordinate denoising frames the generation task as denoising observed coordinates. Single-frame-to-trajectory alignment maps one starting conformer to a complete trajectory. Physical grounding injects physics-aware constraints that improve generated output and the information extractable from learned fingerprints.
RMSF correlations across test sets including high-flexibility challenge
Across the primary test set generated trajectories achieve RMSF correlation of 0.875. On the high-flexibility challenge set the correlation is 0.766. Single-conformer predictions match MD-derived dynamics with comparable agreement.
Single-conformer predictions and agreement with MD-derived dynamics
Predictions from a single RNA conformer show agreement with molecular dynamics-derived dynamics comparable to full-trajectory generation. This demonstrates that the model captures essential dynamic patterns from limited input.
Limitations of the RNADynBench scope and trajectories
The benchmark covers a defined set of RNA families and may not encompass all structural classes or cellular contexts. Real cellular environments include molecular crowding, protein binding, and chemical modifications that are not represented in the current trajectories.