CD-HIT 100 · SINGLE-CHAIN RNA
RNA Structure Prediction Leaderboard
Compare structural accuracy, local atomic quality, and steric clashes under one evaluation protocol.
FULL RESULTS
Method leaderboard
Loading…Full-cohort mean time; hardware and concurrency vary
| Rank | Method | Inputs | Coverage | Mean time |
|---|
Performance overview
One dataset. Multiple perspectives.RANKING
RNA TM-score ranking
↑Higher is better
OBSERVED RUNTIME
Accuracy & prediction time
Mean prediction time in seconds (log scale). Observed hardware and concurrency vary; this is a descriptive comparison.
About the selected metric
- Direction
- Range
- Unit
FEATURED METHOD
MetaFold-RNA3d
MetaFold-RNA3d is our retrained RNA structure prediction model. It uses an architecture similar to AlphaFold 3 and integrates MSA, secondary-structure predictions from MetaFold-RNA, and templates for RNA structure prediction.
TARGET × METHOD
Per-target performance matrix
WeakerBetter
Summarizing target data…
EVALUATION
One protocol, four complementary metrics
The leaderboard reports global topology, coordinate deviation, local atomic environment, and geometric clashes. Mean performance is the default to avoid overemphasizing the best sample.
SCHEMA
—Statistical views
Mean performance averages candidates within each target before aggregating targets. Best candidate may select a different candidate for each metric on a target.
Dataset construction
Candidate RNA target sequences are clustered with CD-HIT at 100% sequence identity. Exact duplicate entries are removed before prediction and evaluation.
Shared MSA and input transparency
To maintain consistent evaluation inputs, methods that use MSA receive the same multiple sequence alignment (MSA) input for each RNA target. MSA, secondary-structure, and template usage are indicated in the results; per-target input records are retained in the source JSON.
DEFINITIONS