RNADyn: A Benchmark for Generating and Understanding RNA Dynamics

Yiming Huang; Lennart Bastian; Hanqun Cao; Luis Vollmers; Tolga Birdal

cs.LGOct 2, 2026

RNADyn: A Benchmark for Generating and Understanding RNA Dynamics

Authors: Yiming Huang, Lennart Bastian, Hanqun Cao, Luis Vollmers, Tolga Birdal

Organizations: Imperial College London · The Chinese University of Hong Kong · Technical University of Munich

Abstract

Ribonucleic acid (RNA) functions through conformational changes that are not fully captured by static structures. However, large-scale standardized RNA dynamics data remain limited, and existing approaches typically treat trajectory generation and dynamics understanding as separate objectives. Here, we introduce RNADynBench, a standardized RNA molecular dynamics (MD) benchmark with 2585 quality-controlled 100-ns all-atom trajectories and leakage-controlled splits. Building on RNADynBench, we develop RNADynNet, a unified model for RNA dynamics learning that uses a shared backbone for both trajectory generation and dynamics fingerprint extraction from a single conformer. It combines coordinate denoising, single-frame-to-trajectory alignment, and physical grounding to connect all-atom trajectory generation with dynamics representation learning. Physical grounding improves both generated dynamics and the physical information recoverable from these fingerprints. Across both test sets, including the high-flexibility challenge set, the generated trajectories achieve RMSF correlations of 0.875 and 0.766, while single-conformer predictions show comparable agreement with MD-derived dynamics. RNADynBench and RNADynNet together establish a benchmark and unified modeling framework for generating and understanding RNA dynamics.

Figures & tables

Figure 1: RNADynBench visualized. a , Mean

R_{\mathrm{g}}

versus RNA residue count. Points show individual RNAs, lines connect length-binned medians, and the inset shows counts;

\rho

is Spearman correlation between RNA length and mean

R_{\mathrm{g}}

. b , Median phosphate–water radial distribution functions (RDFs); shading shows the interquartile range, and vertical lines mark the first peak and minima. c , Weighted median lag-RMSD for size-matched RNA–protein comparisons; shading shows the weighted interquartile range (IQR). d , e , Weighted distributions of mean local RMSF ( d ) and median pairwise RMSD ( e ). Split violins show protein left and matched RNA right, with median, IQR, and 5th–95th percentiles. RNA uses C4 ′ atoms and protein uses C

\alpha

atoms.

Resource	Scale	Simulation time	Task
BIGNASim (2016) \citep BIGNASim-NAR2016	14 traj	15.6 $\mu$ s total	Database and analysis
CASP15 MD \citep CASP15MD-CSBJ2025	9 RNAs / 61 traj	300 ns each (18.3 $\mu$ s total)	Structure refinement
DeepRMSF \citep DeepRMSF-BIB2026	371 RNAs	200 ns each (74.2 $\mu$ s total)	RMSF prediction
RNADynBench	1,469 RNAs / 2,585 traj	100 ns each ( 258.5 $\mu$ s total)	MD generation / prediction

Table 1: Comparison of RNA MD resources.

Figure 2: Overview of RNADynNet. RNADynNet jointly learns trajectory generation and single-conformer dynamics representations from MD trajectories. During training, the trajectory view (blue) provides temporal context for denoising and alignment, while the single-conformer view (pink) learns a dynamics representation from one conformer. Physical grounding supervises both views with MD-derived dynamical statistics. At inference, a single conformer supports two modes: the denoising path generates a trajectory, whereas the encoder readout directly yields a dynamics fingerprint. ST denotes spatiotemporal.

Method	test_struct							test_flex
Method	Bond MAE $\downarrow$	O3P max $\downarrow$	Chiral flip $\downarrow$	Clash /1k $\downarrow$	RMSF $r\uparrow$	PCA $W_{2}\downarrow$	GPU time	Bond MAE $\downarrow$	O3P max $\downarrow$	Chiral flip $\downarrow$	Clash /1k $\downarrow$	RMSF $r\uparrow$	PCA $W_{2}\downarrow$	GPU time
Reference	0.03308	0.1938	0	0	0.9819	0.7295	$8.1\mathrm{h}$	0.03307	0.1966	0	0	0.9633	3.1137	$13.9\mathrm{h}$
ConfRover	0.19315	120.4747	0.03103	581.4520	0.0666	9.3105	340.4s	0.16506	133.0420	0	419.7848	0.2288	11.1922	501.5s
MDGen	0.08024	15.1488	0.03103	95.9398	0.3737	2.4200	7.2s	0.10786	18.9945	$\mathbf{<10^{-5}}$	88.9058	0.2994	8.9282	11.0s
BioKinema	0.02612	4.4682	0.03836	2.0746	0.8770	1.9276	40.2s	0.02950	8.3586	0.02012	5.8241	0.7536	5.9406	47.8s
w/o PG	0.02418	2.4455	0.03097	1.1884	0.8594	1.9085	/	0.02555	9.1171	0.00160	2.8416	0.7440	6.1744	/

Table 2: Trajectory generation on RNADynBench. Bold marks the best.

Figure 3: Dynamics inference from single conformers. Predicted per-nucleotide RMSF and pairwise nucleotide motion couplings from RNADynNet are compared with MD references for two RNAs. Motion-coupling matrices show RNADynNet below the diagonal and MD above; structures are colored by RMSF.

	test_struct							test_flex
	RMSF		Covariance			NMC		RMSF		Covariance			NMC
Method	$r\uparrow$	MAE $\downarrow$	RMWD $\downarrow$	Sym. KL $\downarrow$	IoU $\uparrow$	$r\uparrow$	MAE $\downarrow$	$r\uparrow$	MAE $\downarrow$	RMWD $\downarrow$	Sym. KL $\downarrow$	IoU $\uparrow$	$r\uparrow$	MAE $\downarrow$
MD reference	0.8755	0.3770	0.7960	0.5525	0.5497	0.8532	0.1462	0.7584	1.6415	3.5505	1.3471	0.4185	0.7520	0.2145
RNA-FM †	0.2939	1.6188	2.1173	4.9868	0.1731	0.3690	0.2644	0.2085	3.8148	5.0390	4.2367	0.2318	0.3635	0.2952
Geometry-GNN	0.2801	1.1241	1.5796	2.8662	0.2561	0.4821	0.2591	0.3385	3.4276	4.6870	3.1959	0.3054	0.4442	0.2910
RNA-FM+2D GCN	0.1257	2.0117	2.6303	8.5914	0.1831	0.4797	0.2569	0.1536	3.5511	4.8377	4.9398	0.2625	0.5333	0.2609

Table 3: Single-conformer prediction of MD-derived RNA dynamics. The MD reference compares two non-overlapping 50-ns MD blocks; model predictions are evaluated against full 100-ns MD statistics.

\dagger

denotes probes on frozen representations. Bold denotes the best result.

Figure 4: Predictive data scaling on test_struct . NMC MAE ( a ) and covariance-ellipsoid IoU ( b ) across 10–100% of the training data and 1k–6k training steps.

	AUROC-Biosensor				AUROC-Robin
Method	Random	RNA homology	Ligand fingerprint	Homol. & fingerpr.	Random	RNA homology	Ligand fingerprint	Homol. & fingerpr.
RSAPred \citep RSAPred-BIB2024	0.8764	0.7550	0.6707	0.6019	0.6327	0.5392	0.6320	0.4938
DeepDTIs \citep DeepDTIs-JPR2017	0.9300	0.8399	0.6813	0.6118	0.6302	0.5503	0.6290	0.4987
DeepConv-DTI \citep DeepConvDTI-PLoSCB2019	0.9249	0.8427	0.6894	0.6213	0.6301	0.5625	0.6390	0.5104
GraphDTA \citep GraphDTA-Bioinformatics2021	0.8992	0.8284	0.7014	0.6370	0.6590	0.5528	0.6481	0.5510
GerNA-Bind \citep GerNABind-NMI2025	0.9755	0.9014	0.7723	0.7279	0.7094	0.6188	0.6814	0.6176

Table 4: RNA–ligand binding AUROC ( best , second best ).

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Setting	Value
Simulation engine	GROMACS \citep gromacs-2015
RNA force field	Amber RNA OL3 \citep AmberRNAOL3-2011
Solvent and ions	OPC water \citep OPCwater-2014; Li–Merz 12-6 OPC Na + /Cl - parameters \citep Sengupta-2021; neutralizing ions plus 150 mM NaCl
Periodic box	Rhombic dodecahedron; 12 Å default padding, increased to 18 or 24 Å for periodic-image contacts
Electrostatics	Particle-mesh Ewald \citep PME-1993
Integration	4 fs time step with hydrogen mass repartitioning \citep Feenstra-1999

Appendix

Table 5: RNADynBench molecular dynamics protocol.

Partition	Trajectories	Primary role
Training	2,248	Model optimization and downstream probe training
Validation	102	Checkpoint and hyperparameter selection
test_struct	117	Structural and molecular generalization
test_flex	100	High-flexibility generalization
Homology guard	18	Excluded homology/cluster guards
Total	2,585

Appendix

Table 6: RNADynBench dataset splits.

Figure 5: Size-stratified RNA–protein dynamics. Distributions of ( a ) mean marker-atom RMSF, ( b ) mean RMSD to the first sampled frame, and ( c ) heavy-atom

R_{\mathrm{g}}

coefficient of variation across five size bins. RNA uses C4 ′ markers and protein uses C

\alpha

; vertical axes are logarithmic. Split violins show proteins on the left and matched RNAs on the right; bars indicate the median, interquartile range, and 5th–95th percentiles.

Symbol	Meaning
$L,A,N,d$	Numbers of frames, atoms, residues, and feature dimensions.
$\mathbf{X},\mathbf{x}_{f}$	All-atom trajectory window and its frame- $f$ conformer.
$\theta,\bar{\theta}$	Online parameters and their EMA; fingerprints use the latter.
$\mathcal{D}_{\theta},\mathcal{E}_{\theta},\mathcal{E}_{\bar{\theta}},\ell_{\mathrm{fp}}$	Trajectory denoiser, online single-conformer encoder, EMA fingerprint encoder, and readout block.
$\widetilde{\mathbf{X}},\bm{\sigma},\bm{\epsilon}$	Noised trajectory, coordinate-noise scale tensor, and Gaussian perturbation.
$\mathbf{R}_{\mathrm{traj}},\mathbf{R}_{\mathrm{sf}}$	Independent global rotations of the trajectory and single-frame views.

Appendix

Table 7: Notation used throughout our manuscript.

Model	test_struct						test_flex
Model	Bond MAE $\downarrow$	O3P max $\downarrow$	Chiral flip $\downarrow$	Clash /1k $\downarrow$	RMSF $r\uparrow$	PCA $W_{2}\downarrow$	Bond MAE $\downarrow$	O3P max $\downarrow$	Chiral flip $\downarrow$	Clash /1k $\downarrow$	RMSF $r\uparrow$	PCA $W_{2}\downarrow$
Cov-only	0.02423	1.7091	0.03092	0.9583	0.8712	1.8699	0.02561	3.8573	0.00126	3.2085	0.7609	5.8609
Coupling-only	0.02414	1.0845	0.03088	0.9891	0.8631	1.8983	0.02537	5.4953	0.00093	2.8008	0.7623	5.8490
Full model (Block 24)	0.02430	1.2558	0.03091	0.9256	0.8720	1.8896	0.02586	6.2126	0.00043	3.1947	0.7592	5.8033
Full model (Block 17)	0.02424	1.8858	0.03085	0.8225	0.8746	1.8739	0.02567	4.2400	0.00093	3.0404	0.7658	5.9169

Appendix

Table 8: Trajectory-generation ablations. RNA-group macro means; distances are in Å and chiral flips in percent.

	test_struct							test_flex
	RMSF		Covariance			NMC		RMSF		Covariance			NMC
Method	$r\uparrow$	MAE $\downarrow$	RMWD $\downarrow$	Sym. KL $\downarrow$	IoU $\uparrow$	$r\uparrow$	MAE $\downarrow$	$r\uparrow$	MAE $\downarrow$	RMWD $\downarrow$	Sym. KL $\downarrow$	IoU $\uparrow$	$r\uparrow$	MAE $\downarrow$
Cov-only	0.8540	0.4307	0.8856	0.6007	0.5309	0.7936	0.1722	0.7711	1.7128	3.3259	0.8238	0.4547	0.7537	0.2090
Coupling-only	0.8448	0.4535	0.9297	0.6690	0.4806	0.8160	0.1626	0.7468	1.7138	3.3845	0.9596	0.4322	0.7646	0.2012
Full model (Block 24)	0.8622	0.4124	0.8716	0.5892	0.5276	0.8016	0.1659	0.7696	1.5995	3.2786	0.8498	0.4476	0.7592	0.2023
Full model (Block 17)	0.8626	0.4076	0.8741	0.5823	0.5276	0.7958	0.1694	0.7782	1.7560	3.3359	0.8775	0.4522	0.7577	0.2035

Appendix

Table 9: Single-conformer dynamics ablations using independent probes on frozen EMA backbones. RNA-group macro means; RMWD and RMSF MAE are in Å.

Figure 6: Predictive data scaling. Orange curves darken from 1k to 6k training updates across 10–100% of the training data. Shading shows pointwise 95% source-RNA-group bootstrap intervals for 6k only. Each checkpoint uses its EMA backbone and EMA descriptor heads.

Figure 7: Stepwise denoising of the low-motion held-out RNA 3IVN_A. Columns show 0, 25, 50, 75, and 100 ns. Step rows are diagnostic $\hat{\mathbf{x}}_{0}$ estimates at the labeled noise scales; the final two rows show the generated trajectory and reference MD. Ribbon gaps mark O3 ′ –P distances of at least 2.6 Å.

Figure 8: Stepwise denoising of the median-motion held-out RNA 3SKR_A. Columns show 0, 25, 50, 75, and 100 ns. Step rows are diagnostic $\hat{\mathbf{x}}_{0}$ estimates at the labeled noise scales; the final two rows show the generated trajectory and reference MD. Ribbon gaps mark O3 ′ –P distances of at least 2.6 Å.

Figure 9: Stepwise denoising of the high-motion held-out RNA 6AZ4_A. Columns show 0, 25, 50, 75, and 100 ns. Step rows are diagnostic $\hat{\mathbf{x}}_{0}$ estimates at the labeled noise scales; the final two rows show the generated trajectory and reference MD. Ribbon gaps mark O3 ′ –P distances of at least 2.6 Å.

Molecular Dynamics Simulation

Molecular dynamics simulation models atomic motion to study molecular structure, material behavior, reactions, and thermodynamic properties.

See all 55 papers

mdRNAcat: A Molecular Dynamics RNA Catalog of REST2 Conformational Ensembles
Oct 26, 2026Anton E. Dorn, Emile de Bruyn, Fabrice von der Lehr +3Molecular Dynamics Simulation
Jülich Supercomputing Centre, Forschungszentrum Jülich (FZJ) · Faculty of Physics, Georg August University of Göttingen · Institute of Software Technology, Germany Aerospace Center (DLR) +2
RiboSphere: Learning Unified and Efficient Representations of RNA Structures
Mar 20, 2026Zhou Zhang, Hanqun Cao, Cheng Tan +3Molecular Representation Learning Representation Learning
State Key Laboratory for Novel Software Technology at Nanjing University, School of Computer Science, Nanjing University · The Chinese University of Hong Kong · Stanford University
REFLEX: RNA Ensemble Generation via Flexibility-Calibrated Stochastic Bridge
Oct 19, 2026Wanli Ma, Tianmeng Hu, Ke LiMolecular Generation Bioinformatics
University of Exeter · Earlham Institute

Oct 26, 2026J/K move · Enter open · S save

mdRNAcat: A Molecular Dynamics RNA Catalog of REST2 Conformational Ensembles

Anton E. Dorn, Emile de Bruyn, Fabrice von der Lehr, Philipp Knechtges+2

Jülich Supercomputing Centre, Forschungszentrum Jülich (FZJ) · Faculty of Physics, Georg August University of Göttingen · Institute of Software Technology, Germany Aerospace Center (DLR)+2

Molecular Dynamics Simulation

Open paper AskPDF arXiv

Learning molecular models of RNA requires simulation data that capture conformational variability and provide consistent physical training signals. Here, we introduce mdRNAcat, a curated dataset of all-atom molecular dynamics (MD) simulations of 48 RNA systems selected to promote sequence diversity. Each system is simulated using replica-exchange with solute scaling (REST2), with 32 Hamiltonian states and production lengths of 200-300 ns per state. The dataset provides atomic coordinates, instantaneous forces, and forces reevaluated under the unscaled reference Hamiltonian, together with structural annotations and replica-exchange metadata. Coordinates and forces are available at 1 ps intervals for the reference state and 100 ps intervals across all states in a standardized HDF5 format. We characterize conformational sampling, exchange behavior, base pairing, and sugar-pucker distributions to document the properties of the simulated ensembles. mdRNAcat provides a reusable resource for developing and evaluating data-driven RNA models, including coarse-grained force fields.

Abstract

Molecular Dynamics Simulation

Explore similar work

mdRNAcat: A Molecular Dynamics RNA Catalog of REST2 Conformational Ensembles

RiboSphere: Learning Unified and Efficient Representations of RNA Structures

REFLEX: RNA Ensemble Generation via Flexibility-Calibrated Stochastic Bridge