RNA Secondary Structure Prediction — MFE Fold (Zuker / Turner)
Fold an RNA to its minimum-free-energy secondary structure — dot-bracket, MFE, base-pair list and an exportable arc diagram, with nothing to install.
Predicted, not measured
- How good is it?
- Single-sequence folding of this kind recovers roughly 40-70% of the base pairs in a known structure, and gets worse as the sequence gets longer. What comes back is also one structure out of many: another within a fraction of a kcal/mol of it is common, so use this to compare candidates or to ask whether a region tends to be structured at all, not as the structure.
- Only valid for:
- a single strand up to 3,000 nt at a fixed 37 °C (600 nt when folded in the workbench, which runs the fallback engine in the browser). No pseudoknots, no two-strand hybridization — use oligo_cofold for that — and no temperature dependence, so a structure predicted here is not a structure at your annealing temperature.
How this was measured, and what it was fitted on
Benchmarks of MFE folding against known structures put single-sequence accuracy near 40-70% of base pairs recovered, falling as the sequence gets longer; no held-out statistic is measured here. The MFE structure is one structure out of an ensemble, and a second structure within a fraction of a kcal/mol of it is common.
Turner 2004 nearest-neighbor free energies from optical melting experiments, as distributed with ViennaRNA. The fallback engine used only where the folding service is unconfigured is this repo's own Zuker DP, whose loop-initiation tables are a documented hybrid of Turner 1999 and 2004; it reports itself in `engine` and its energies are NOT on the same scale.
Paste an RNA sequence (or DNA — it's treated as RNA) and fold it to its minimum-free-energy secondary structure, up to 3,000 nt. It folds on the Turner 2004 nearest-neighbor energies at 37 °C — the same answer a local RNAfold gives, with nothing to install. You get the dot-bracket structure, the MFE in kcal/mol, the base-pair list, and an arc diagram you can export. What it does not do is as important: it predicts one MFE structure out of an ensemble that often holds near-ties within a fraction of a kcal/mol, it allows no pseudoknots, and it folds only at 37 °C — so a structure here is not a structure at your annealing temperature.
Design, edit, and verify complete constructsSeqStudio combines sequence editing and annotation, plasmid maps, primer/cloning/CRISPR design, Sanger verification, and GenBank/SnapGene files in one workspace.
How to use the RNA Structure (MFE) tool
- 1Paste an RNA/DNA sequence (up to 3,000 nt) or load the tRNA example.
- 2Read the MFE, base-pair count and dot-bracket structure, and inspect the arc diagram (each arc is a base pair).
- 3Export the arc diagram, or copy the dot-bracket — and remember an MFE structure is one member of an ensemble, so compare candidates rather than trusting a single fold.
Frequently asked questions
What algorithm and energies does it use?
A minimum-free-energy fold on the full Turner 2004 nearest-neighbor loop model at 37 °C. It predicts one MFE structure and does not allow pseudoknots, which is standard for this class of tool.
How accurate is the MFE?
The energies are ViennaRNA's own, so they are directly comparable with a local RNAfold run on the same parameters. The limit is the method, not the implementation: benchmarks of single-sequence MFE folding against known structures recover roughly 40-70% of base pairs, and accuracy falls as the sequence gets longer. The MFE structure is also only the lowest-energy member of an ensemble that frequently contains near-ties, so it is strongest for ranking candidates and for asking whether a region tends to be structured at all.
Why not just use ViennaRNA?
The structure and the energy match what RNAfold gives you locally, on the same Turner 2004 parameters. What a local install adds is everything this page deliberately does not expose: the partition function and base-pair probabilities, folding at other temperatures, constraint folding, suboptimal structures, and sequences past 3,000 nt. Use this when you want one fold now from a browser or an API call; install RNAfold when you need the rest of it.
Is there a size limit?
Yes — 3,000 nt, because folding time grows with the cube of the length. Fold a specific region — a 5' UTR, a terminator, a riboswitch — rather than a whole transcript.
Can I use this from code?
Yes — the rna_fold tool is available via the REST API and MCP server.