Evaluation / Manuscript results

Benchmark results

Two dynamics tasks, complementary test splits. Binding transfer is reported separately.

These tables reproduce a local manuscript snapshot. Public checkpoint, evaluator version and runnable evaluation package: Coming soon. No community ranking is implied.

Trajectory generation Single-conformer prediction Binding transfer

Trajectory generation

Input: an initial conformer. Output: 101 coordinate frames over 100 ns. Geometry metrics use future 100 frames; RMSF uses all 101. Aggregate within RNA first, then across RNAs.

test_struct

Method	K	Bond MAE ↓	O3′–P max ↓	Chiral flip (%) ↓	Clash / 1k ↓	RMSF r ↑	PCA W₂ ↓	Time / trajectory
Reference	—	0.03308	0.1938	0	0	0.9819	0.7295	8.1 h
ConfRover	1	0.19315	120.4747	0.03103	581.4520	0.0666	9.3105	340.4s
MDGen	4	0.08024	15.1488	0.03103	95.9398	0.3737	2.4200	7.2s
BioKinema	4	0.02612	4.4682	0.03836	2.0746	0.8770	1.9276	40.2s
w/o PG	4	0.02418	2.4455	0.03097	1.1884	0.8594	1.9085	/
RNADynNet	4	0.02424	1.8858	0.03085	0.8225	0.8746	1.8739	24.5s

test_flex

Method	K	Bond MAE ↓	O3′–P max ↓	Chiral flip (%) ↓	Clash / 1k ↓	RMSF r ↑	PCA W₂ ↓	Time / trajectory
Reference	—	0.03307	0.1966	0	0	0.9633	3.1137	13.9h
ConfRover	1	0.16506	133.0420	0	419.7848	0.2288	11.1922	501.5s
MDGen	4	0.10786	18.9945	mathbf{<10^{-5	88.9058	0.2994	8.9282	11.0s
BioKinema	4	0.02950	8.3586	0.02012	5.8241	0.7536	5.9406	47.8s
w/o PG	4	0.02555	9.1171	0.00160	2.8416	0.7440	6.1744	/
RNADynNet	4	0.02567	4.2400	0.00093	3.0404	0.7658	5.9169	32.9s

Distances in Å. Generation runtimes measured per trajectory on H200. Reference rows report MD quantities.

Single-conformer dynamics prediction

Input: one conformer. Output: per-residue covariance, RMSF and normalized motion coupling (NMC). † identifies independently trained probes on frozen representations; other learned-model rows use direct predictors.

test_struct

Method	RMSF r ↑	RMSF MAE ↓	Cov. RMWD ↓	Cov. sym. KL ↓	Cov. IoU ↑	NMC r ↑	NMC MAE ↓
MD reference	0.8755	0.3770	0.7960	0.5525	0.5497	0.8532	0.1462
RNA-FM †	0.2939	1.6188	2.1173	4.9868	0.1731	0.3690	0.2644
Geometry-GNN	0.2801	1.1241	1.5796	2.8662	0.2561	0.4821	0.2591
RNA-FM+2D GCN	0.1257	2.0117	2.6303	8.5914	0.1831	0.4797	0.2569
BackFlip	0.8098	0.4608	0.9449	0.7599	0.4919	0.7948	0.1707
DynaProt	0.7726	0.5266	0.9786	0.8288	0.4502	0.2454	0.2775
w/o align. and PG †	0.7473	0.6056	1.0756	0.9827	0.3966	0.7924	0.1717
w/o PG †	0.7797	0.5160	0.9839	0.7542	0.4540	0.7834	0.1744
RNADynNet †	0.8626	0.4076	0.8741	0.5823	0.5276	0.7958	0.1694
RNADynNet	0.8665	0.3983	0.8651	0.5743	0.5260	0.8115	0.1629

test_flex

Method	RMSF r ↑	RMSF MAE ↓	Cov. RMWD ↓	Cov. sym. KL ↓	Cov. IoU ↑	NMC r ↑	NMC MAE ↓
MD reference	0.7584	1.6415	3.5505	1.3471	0.4185	0.7520	0.2145
RNA-FM †	0.2085	3.8148	5.0390	4.2367	0.2318	0.3635	0.2952
Geometry-GNN	0.3385	3.4276	4.6870	3.1959	0.3054	0.4442	0.2910
RNA-FM+2D GCN	0.1536	3.5511	4.8377	4.9398	0.2625	0.5333	0.2609
BackFlip	0.7311	1.7974	3.7118	4.4011	0.4058	0.7399	0.2084
DynaProt	0.5518	2.5861	3.9757	1.9398	0.3692	0.1889	0.3132
w/o align. and PG †	0.6232	2.1515	3.8137	1.3343	0.3811	0.7506	0.2088
w/o PG †	0.6920	1.8983	3.5810	1.1055	0.4069	0.7443	0.2101
RNADynNet †	0.7782	1.7560	3.3359	0.8775	0.4522	0.7577	0.2035
RNADynNet	0.7827	1.5800	3.2700	0.8533	0.4528	0.7614	0.2027

MD reference compares two non-overlapping 50-ns blocks; model predictions are evaluated against full 100-ns statistics. Native-head and frozen-probe results are distinct.

RNA–ligand binding transfer

RNADynNet exceeds GerNA-Bind in five of eight reported settings. Improvements are not universal.

Method	Biosensor · Random	Biosensor · RNA homology	Biosensor · Ligand fingerprint	Biosensor · Both	Robin · Random	Robin · RNA homology	Robin · Ligand fingerprint	Robin · Both
RSAPred	0.8764	0.7550	0.6707	0.6019	0.6327	0.5392	0.6320	0.4938
DeepDTIs	0.9300	0.8399	0.6813	0.6118	0.6302	0.5503	0.6290	0.4987
DeepConv-DTI	0.9249	0.8427	0.6894	0.6213	0.6301	0.5625	0.6390	0.5104
GraphDTA	0.8992	0.8284	0.7014	0.6370	0.6590	0.5528	0.6481	0.5510
GerNA-Bind	0.9755	0.9014	0.7723	0.7279	0.7094	0.6188	0.6814	0.6176
RNADynNet	0.9582	0.9066	0.7041	0.7673	0.7119	0.6899	0.6760	0.6619

Reproduce the evaluation

Evaluator code, environment, checkpoint identities and runnable examples: Coming soon.

Read the evaluation contract →