mdRNAcat: A Molecular Dynamics RNA Catalog of REST2 Conformational Ensembles

Anton E. Dorn; Emile de Bruyn; Fabrice von der Lehr; Philipp Knechtges; Stefan Kesselheim; Alexander Schug

q-bio.BMOct 26, 2026

mdRNAcat: A Molecular Dynamics RNA Catalog of REST2 Conformational Ensembles

Authors: Anton E. Dorn, Emile de Bruyn, Fabrice von der Lehr, Philipp Knechtges, Stefan Kesselheim, Alexander Schug

Organizations: Jülich Supercomputing Centre, Forschungszentrum Jülich (FZJ) · Faculty of Physics, Georg August University of Göttingen · Institute of Software Technology, Germany Aerospace Center (DLR) · Department of Physics, University of Cologne ·

Abstract

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.

Figures & tables

Figure 1: ( a ) Distribution of the number of residues per RNA system in the dataset. ( b ) Distribution of the percentage of paired residues in the dataset; most systems display similar behavior, apart from one RNA system in an extended state. Colors indicate different RNA domains.

Total RNA systems	$48$
Systems with $200\text{\,}\mathrm{ns}$ simulation time	$45$
Systems with $300\text{\,}\mathrm{ns}$ simulation time	$3$
Replicas	$32$
State trajectories	$1,536$
Total sampled time	$316.8$	$\mathrm{\SIUnitSymbolMicro s}$
Total sampled time at $300\text{\,}\mathrm{K}$	$9.9$	$\mathrm{\SIUnitSymbolMicro s}$

Table 1: Statistical overview of the dataset.

Field	Size	Type	Unit	Description
<pdbid>/
chain	$N$	string		Chain ID
element	$N$	string		Chemical element
resid	$N$	integer		Residue number
resname	$N$	string		Residue name
z	$N$	integer		Atomic number

Table 2: mdRNAcat is made available as HDF5 files with an hierarchical organization as shown in this table. For each system in the data set, there are 2 files both labeled with the respective PDB ID. The first contains only the lowest trajectory with

1\text{\,}\mathrm{ps}

time steps. The second file contains all replicas and additional derived metrics indicated with a blue background in the table with

100\text{\,}\mathrm{ps}

time steps. Key: <T> , REST2 effective temperature (K);

N

, number of atoms;

R

, number of residues;

F

, trajectory length in frames (

=\,

frameStridePs );

M

, number of replicas;

K

, energy terms;

S

, energy samples;

X

, exchange records;

P

, log segments. 2 2 2 A comment on the units used in mdRNAcat can be found in the Appendix A.1

Figure 2: Sequence and structural-motif coverage of mdRNAcat against filtered X-ray PDB reference sets. ( a ) Coverage versus locally normalized nearest-representative distance. ( b , c ) UMAP visualizations McInnes et al. (2018a) ; McInnes et al. (2018b) ; stars denote mdRNAcat representatives and colors show smoothed distance ratios on separate scales, with structural values clipped for display. ( d ) Mean structural-descriptor differences in units of the reference-set standard deviation. Coverage is calculated before UMAP projection.

Figure 3: Free energy landscape (FEL) of the lowest temperature ( $300\text{\,}\mathrm{K}$ ) trajectory of the 6qn3 system, the axes are determined with time-lagged independent component analysis (TICA) ( Pérez-Hernández et al. (2013) ), marked with the experimentally resolved structure (exp) and the two lowest minima (1 and 2); experimentally resolved structure is displayed to the left and structures from the two minima to the right; in yellow FEL of a plain single replica $300\text{\,}\mathrm{K}$ simulation of 6qn3 (not part of mdRNAcat) mapped to the same TICA coordinates.

Figure 4: ( a ) Transition probability between two thermodynamic states and energy overlap between two thermodynamic states against the 31 transitions of the states for the 3r4f system, with dotted lines indicating the average over all transitions. ( b ) Density of the thermodynamic states against the energy of the same system. ( c ) Density of the replicas against the energy of the same system.

Avg. transition probability	$0.183$	$\pm 0.065$
Avg. energy overlap	$0.269$	$\pm 0.070$

Table 3: Replica exchange statistics averaged over all 48 systems with standard deviation between systems.

Figure 5: Free energy landscape (FEL) of the 2qus, 6cu1 and 7elq systems, the axes are determined with time-lagged independent component analysis (TICA) ( Pérez-Hernández et al. (2013) ) at

300\text{\,}\mathrm{K}

354\text{\,}\mathrm{K}

424\text{\,}\mathrm{K}

and

500\text{\,}\mathrm{K}

. Note that the axes are capped at the extension of the

424\text{\,}\mathrm{K}

-plot, so the

500\text{\,}\mathrm{K}

-plot has values outside of that range not displayed. The conformation space gets explored more with higher temperature.

Figure 6: ( a ) Wheel histogram over the sugar-pucker angle of all residues of all systems; north (

0\text{\,}\mathrm{\SIUnitSymbolDegree}

) indicates C3’-endo, so A-form helical pucker; south (

180\text{\,}\mathrm{\SIUnitSymbolDegree}

) indicates C2’-endo, so loops and bulges. ( b ) Distribution of the residue pair likelihood against the sequence separation between the two paired residue for all systems in the data set; thick is the set mean. ( c ) Same as ( b ) but the x-axis normalized to 1.

Figure 7: Probability of a paired residue in residue and temperature dimensions for the 1xjr, 6lax, 7elq and 9obm systems. On the left of each plot is a bow plot indicating the paired residues in the experimentally resolved structure.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

PDB ID	Residues	Simulation length	Status
1duh	44	$200\text{\,}\mathrm{ns}$	Part of mdRNAcat
1ehz	76		Modified residues not parameterized
1hr2	158		Too large
1kh6	53		Modified residues not parameterized
1kxk	70	$300\text{\,}\mathrm{ns}$	Part of mdRNAcat
1nbs	155		Missing residues in the pdb

Appendix

Table 4: RNA systems considered or used in mdRNAcat.

Parameter	Value
integrator	steep
nsteps	100000
emtol	10
emstep	0.01
nstlog	1000
nstenergy	100

Appendix

Table 5: Parameters used for energy minimization.

Parameter	Value
define	-DPOSRES
integrator	md
nsteps	50000
dt	0.002
nstxout	500
nstvout	500

Appendix

Table 6: Parameters used in the NVT equilibration.

Parameter	Value
define	-DPOSRES
integrator	md
nsteps	50000
dt	0.002
nstxout	500
nstvout	500

Appendix

Table 7: Parameters used for NPT equilibration.

Parameter	Value
integrator	md
nsteps	100000000
dt	0.002
nstxout	250
nstvout	250
nstfout	250

Appendix

Table 8: Parameters used for production simulation runs.

Replica	$T_{i}$ [ $\mathrm{K}$ ]	$\lambda$	$\sqrt{\lambda}$
1	300.000000	1.000000	1.000000
2	304.980000	0.983700	0.991817
3	310.050000	0.967600	0.983667
4	315.200000	0.951800	0.975602
5	320.440000	0.936200	0.967574
6	325.760000	0.920900	0.959635

Appendix

Table 9: REST2 effective-temperature ladder and

\lambda

scaling factors.

Figure 8: Density of the thermodynamic states against the energy of the first 24 systems in mdRNAcat.

Figure 9: Density of the thermodynamic states against the energy of the second 24 systems in mdRNAcat.

Molecular Dynamics Simulation

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

See all 55 papers

RNADyn: A Benchmark for Generating and Understanding RNA Dynamics
Oct 2, 2026Yiming Huang, Lennart Bastian, Hanqun Cao +2Molecular Dynamics Simulation Molecular Representation Learning
Imperial College London · The Chinese University of Hong Kong · Technical University of Munich
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 2, 2026J/K move · Enter open · S save

RNADyn: A Benchmark for Generating and Understanding RNA Dynamics

Yiming Huang, Lennart Bastian, Hanqun Cao, Luis Vollmers+1

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

Molecular Dynamics Simulation Molecular Representation Learning

Open paper AskPDF arXiv

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.

Abstract

Molecular Dynamics Simulation

Explore similar work

RNADyn: A Benchmark for Generating and Understanding RNA Dynamics

RiboSphere: Learning Unified and Efficient Representations of RNA Structures

REFLEX: RNA Ensemble Generation via Flexibility-Calibrated Stochastic Bridge