How to read the results

RFoldseek encodes every nucleotide as one of 80 learned 3Dr structural states, aligns the resulting token strings, and reports the closest structures in each database.

1. Choose a mode

RFoldseek (recommended for most searches) prefilters with a k-mer index, aligns structural tokens with affine-gap Smith–Waterman, superposes the aligned C4′ atoms, and ranks hits by the structure score S_SW × TM-score. RFoldseek-TM re-scores the same candidates with US-align on the coordinates, which improves the structural ranking at a higher cost.

2. The hit table

Click a column heading to sort by it, and a row to open the hit.

Column	Meaning
Target	The matched chain: the PDB entry with its chain, or the RNAcentral id of a predicted model (it links to the entry page).
RNA / organism	The name of the RNA and its source organism. Predicted models also show their Rfam family when one is assigned.
pLDDT	Predicted databases only: the mean pLDDT of the model, a measure of its quality (higher is better, 100 is the maximum).
Seq. identity	Nucleotide identity over the aligned positions. Databases without sequences show 3Dr identity instead: the fraction of aligned pairs with the same structural token.
RMSD (Å)	Root-mean-square deviation of the superposed aligned nucleotides.
Structure score	The ranking score, SSW × TM-score. A hit has to match locally and in the arrangement of its aligned segments.
TM-score	The TM-score of the token alignment after superposition, normalised by the query and by the target length and averaged. In RFoldseek-TM it is the US-align TM-score of the two whole chains.
E-value	The expected number of unrelated hits at least this good by chance. It is calibrated for each query (its length and token composition) against the database it is searched in, and is computed from SSW, so it does not always increase down the list. Databases that have not been calibrated yet report no E-value.
Query cov., Target cov.	The share of the query and of the target that the alignment spans.

3. Inside a hit

Click a row to open it. The summary line gives the alignment length, the query and target ranges, and the 3Dr identity. Aligned-region TM-score is the TM-score computed over the aligned nucleotides only, so it can differ from the TM-score in the table. For predicted models the expanded hit also shows the MSA depth behind the prediction.

The 3Dr line

Between the query and target sequences, | marks the same structural token, + a positive Structure-BLOSUM substitution and . any other aligned pair. Positions count nucleotides with a complete C3′, C4′, O4′ backbone. Databases without sequences show the target residues as ·.

The 3D view

The query is orange and the hit is blue after superposition on the aligned nucleotides; pale colours are unaligned. Rotate and zoom with the mouse, and click a nucleotide to see which nucleotide of the other structure it is aligned to.

4. Databases

PDB-RNA (all RCSB RNA chains) and RNAsolo 2.0 (a non-redundant representative set) hold experimental structures. NuFold holds predicted structures, which carry no sequence, so 3Dr identity is the main measure of agreement there.