REFLEX: RNA Ensemble Generation via Flexibility-Calibrated Stochastic Bridge

Wanli Ma; Tianmeng Hu; Ke Li

cs.AIOct 19, 2026

REFLEX: RNA Ensemble Generation via Flexibility-Calibrated Stochastic Bridge

Authors: Wanli Ma, Tianmeng Hu, Ke Li

Organizations: University of Exeter · Earlham Institute

Abstract

RNA molecules often function through multiple experimentally observed conformational states. A central challenge is to capture ensemble diversity without sacrificing structural fidelity: samples may collapse toward averaged representative structures or drift into unsupported conformational regions. We introduce REFLEX (RNA Ensemble generation calibrated by FLEXibility), a sequence-conditioned generator for RNA backbone-frame ensembles based on a heteroscedastic stochastic bridge on SE(3). During training, experimental B-factors calibrate residue-wise bridge widths, exposing each observed conformer through local perturbation neighborhoods with residue-specific scales while keeping constrained regions sharper. A learned flexibility condition guides the bridge-field network to recover target conformer-specific frames from these calibrated noisy states. At inference, REFLEX uses only the input sequence, requiring no MSA-subsampling heuristics. On a multi-conformer RNA benchmark, REFLEX achieves the best precision-recall trade-off for ensemble coverage across evaluation thresholds, while recovery-stress ablations indicate gains in target-frame recovery under stronger perturbations.

Figures & tables

Figure 1: Overview of REFLEX . a : RNA ensemble generation aims to recover accurate conformational alternatives rather than simply increase sample spread. b : Flexibility-calibrated stochastic bridge on

\mathrm{SE}(3)

. The structures on the right are multiple target conformations, and their averaged

\mathrm{B}

-factor profile captures nucleotide-wise flexibility used to calibrate the stochastic bridge. c : Nucleotide-wise stochastic strength, with nucleotides on the horizontal axis and stochasticity strength on the vertical axis; flexible positions receive stronger stochasticity than constrained positions.

Figure 2: Coverage of experimental conformers. a : Top-

K

AMR (TM-score), the mean of the

K

highest matching scores for each reference conformer, for

K=1,\ldots,5

, evaluated on clusters where every method has at least five samples. b : Best-match RMSD AMR on the test set. Shading and error bars show

95\%

bootstrap confidence intervals; points in b show per-cluster values.

Figure 3: Qualitative recovery on PDB 8UO6 . Two experimentally resolved conformers are shown against their nearest generated matches from REFLEX , RNAbpFlow , and RhoFold+ subMSA. TM-score and RMSD are reported below each match.

Method	AMR-R	COV-R			AMR-P	COV-P			Diversity (Å)
Method	AMR-R	@ $0.30$	@ $0.35$	@ $0.40$	AMR-P	@ $0.30$	@ $0.35$	@ $0.40$	Diversity (Å)
RNAbpFlow	$0.385$	$0.640$	$0.422$	$0.358$	$0.378$	$0.558$	$0.386$	$0.371$	$3.178$
AF3 subMSA	$0.430$	$0.646$	$0.542$	$0.445$	$0.411$	$0.563$	$0.483$	$0.394$	$0.786$
RhoFold+ subMSA	$0.380$	$0.552$	$0.470$	$0.384$	$0.402$	$0.522$	$0.437$	$0.388$	$1.237$
DynaRNA	$0.331$	$0.436$	$0.327$	$0.247$	$0.257$	$0.250$	$0.218$	$0.204$	$4.439$
REFLEX	$\mathbf{0.451}$	$\mathbf{0.742}$	$\mathbf{0.647}$	$\mathbf{0.543}$	$\mathbf{0.419}$	$\mathbf{0.634}$	$\mathbf{0.533}$	$\mathbf{0.450}$	$2.900$

Table 1: Ensemble recovery on the RNA benchmark. All methods generate

K=20

samples per RNA. AMR-R and AMR-P report reference-to-sample and sample-to-reference best-match TM-scores. COV-R and COV-P use thresholds

0.30

0.35

, and

0.40

. Each metric is macro-averaged over its evaluated clusters (Appendix C ); higher AMR and coverage values are better, with the best in bold . Diversity is average pairwise RMSD in Å.

Figure 4: Precision–recall Pareto plot. Each method is evaluated at TM-score thresholds

\tau\in\{0.30,0.35,0.40\}

. COV-R and COV-P are macro-averaged as in Table 1 ; the gray shaded region denotes the dominated region.

Figure 5: Multi-conformer supervision stability. a : Target-frame recovery error after one-step recovery from perturbed bridge states at

t=0.75

. b : SepRet under the same perturbation protocol. Stars and ns denote significant and non-significant paired comparisons with REFLEX . Additional results at

t=0.25

and

t=0.5

are provided in Appendix D .

Method	Low $[0.078,0.363]$	Mid $[0.365,0.471]$	High $[0.483,0.843]$
Linear Bridge	$0.505$	$0.435$	$0.340$
Uniform Stochastic Bridge	$\mathbf{0.540}$	$0.444$	$0.356$
REFLEX	$0.531$	$\mathbf{0.470}$	$\mathbf{0.358}$

Table 2: Conformer recovery by ground-truth distinctness. Distinctness is average pairwise TM-distance between resolved conformers. The Low/Mid/High bins contain

16/17/17

clusters. Entries are mean per-cluster AMR-R over the evaluated structure sets; higher is better.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Group	Symbol / Name	Value
Time grid	Cutoff $\epsilon_{t}$	$10^{-2}$
	Inference / validation steps	$60$ / $60$
Stochastic bridge	Base widths $\sigma_{x}$ / $\sigma_{\Omega}$	$1.0$ Å / $1.0$ rad
	Heteroscedastic gain	$\sigma_{i}=\sigma(1+2\alpha_{i})$ , with $\alpha_{i}=(\operatorname{clip}(\widetilde{B}_{i},0.2,1.6)-0.2)/1.4$
	Diffusion $g_{x}$ / $g_{R}$	$1.0$ Å / $1.0$ rad; constant
	Rotation schedule	Exponential, $\rho=10$

Appendix

Table 3: Key hyperparameters of REFLEX .

Figure 6: Additional recovery-stress results at

t=0.25

. a : Target-frame recovery error after one-step recovery from perturbed bridge states. b : SepRet under the same perturbation protocol. Stars and ns denote significant and non-significant paired comparisons with REFLEX .

Figure 7: Additional recovery-stress results at

t=0.5

Figure 8: Statistical analysis of conformer recovery. Left: best-match TM-score AMR, where higher is better. Right: best-match RMSD AMR, where lower is better. Bars show means with

95\%

bootstrap confidence intervals; stars denote Holm-corrected significance from one-sided paired Wilcoxon comparisons with REFLEX .

Figure 9: Statistical analysis for precision–recall coverage. Left: COV-R at TM-score thresholds

\tau\in\{0.30,0.35,0.40\}

. Right: COV-P at the same thresholds. Bars show cluster-level means with

95\%

bootstrap confidence intervals; stars denote Holm-corrected significance from one-sided paired bootstrap comparisons with REFLEX . Absence of a star indicates that the comparison does not meet the

0.05

significance threshold.

Variant	COV-R			COV-P			Diversity (Å)
Variant	@ $0.30$	@ $0.35$	@ $0.40$	@ $0.30$	@ $0.35$	@ $0.40$	Diversity (Å)
Linear Bridge	$0.665$	$0.570$	$0.449$	$0.554$	$0.427$	$0.383$	$3.401$
Uniform Stochastic Bridge	$0.687$	$0.601$	$0.434$	$0.606$	$0.466$	$0.404$	$3.088$
w/o flexibility supervision	$\mathbf{0.753}$	$0.616$	$0.452$	$\mathbf{0.647}$	$0.506$	$0.410$	$3.157$
w/o per-residue calibration	$0.717$	$0.620$	$0.436$	$0.620$	$0.471$	$0.401$	$3.252$
REFLEX	$0.742$	$\mathbf{0.647}$	$\mathbf{0.543}$	$0.634$	$\mathbf{0.533}$	$\mathbf{0.450}$	$2.900$

Appendix

Table 4: Matched component ablations. All variants use

K=20

samples per RNA. Coverage is macro-averaged across clusters, with the best value in each column in bold . Diversity is average pairwise RMSD in Å. The two “w/o” rows each remove one component from REFLEX .

Metric	Chain-macro result	$95\%$ CI
Accuracy within $\pm 2$ bins	$0.759$	$[0.691,0.824]$
AUROC for the top $25\%$ by the proxy	$0.681$	$[0.630,0.731]$

Appendix

Table 5: FlexibilityNet prediction. Metrics assess agreement with the relative

\mathrm{B}

-factor-derived proxy.

Method	Precision	Recall	F1
AF3 subMSA	$0.697$	$\mathbf{0.787}$	$0.729$
RNAbpFlow	$0.712$	$0.757$	$0.725$
RhoFold+ subMSA	$0.694$	$0.630$	$0.647$
REFLEX	$\mathbf{0.734}$	$0.765$	$\mathbf{0.741}$

Appendix

Table 6: DSSR base-pair consistency of TM-score-selected matches. Higher values are better; the best observed value in each column is in bold .

Molecular Generation

Molecular generation develops computational methods to create chemical structures and conformations for discovery, optimization, and structure elucidation.

See all 233 papers

RiboBridge: protein-conditioned de novo RNA binder structure design
Oct 22, 2026Zhou Zhang, Hanqun Cao, Minghao Sun +5Molecule Design
State Key Laboratory for Novel Software Technology at Nanjing University, School of Computer Science, Nanjing University · The Chinese University of Hong Kong · National University of Singapore
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
GOLD: Geometric Optimized Latent Diffusion for Structure-Aware RNA Inverse Folding
Oct 16, 2026Qi Si, Xuyang Liu, Penglei Wang +6Molecular Generation Inverse Design
Shanghai Academy of Artificial Intelligence for Science · School of Biomedical Engineering, Shanghai Jiao Tong University · Community Science Department, Kuaishou +2

Oct 22, 2026J/K move · Enter open · S save

RiboBridge: protein-conditioned de novo RNA binder structure design

Zhou Zhang, Hanqun Cao, Minghao Sun, Odin Zhang+4

State Key Laboratory for Novel Software Technology at Nanjing University, School of Computer Science, Nanjing University · The Chinese University of Hong Kong · National University of Singapore

Molecule Design

Open paper AskPDF arXiv

Designing RNA molecules that bind to target protein structures requires generating both an RNA conformation and its bound complex geometry, without assuming a predefined binding site. Existing approaches generally model these coupled processes within a unified representation, entangling RNA conformation generation with protein-relative placement. In this work, we introduce RiboBridge, a structure-first framework that separates RNA conformational generation from protein--RNA assembly through cross-frame interaction geometry. RiboBridge first proposes multiple candidate RNA-binding interfaces on the protein surface, generates canonical RNA conformations within a pretrained RNA structural latent space (RiboSphere), and then predicts protein--RNA interaction constraints to bridge the independently defined RNA and protein coordinate frames. The resulting interaction geometry is converted into a geometrically consistent protein-relative pose through rigid placement. We evaluate RiboBridge on a temporally split, sequence-filtered, and redundancy-controlled benchmark of 124 protein--RNA complexes. Under post-hoc interface-correspondence evaluation, where native information is used only after generation to select among already generated predicted-pocket candidates, RiboBridge achieves the highest fNAT and DockQ and the lowest interface and ligand RMSDs among compared methods. Component ablations and stage-wise diagnostics further characterize the contributions of RNA conformational generation, interaction prediction, and rigid placement, identifying the main sources of complex-recovery error.

Abstract

Molecular Generation

Explore similar work

RiboBridge: protein-conditioned de novo RNA binder structure design

RNADyn: A Benchmark for Generating and Understanding RNA Dynamics

GOLD: Geometric Optimized Latent Diffusion for Structure-Aware RNA Inverse Folding