SeqBench

RNA Secondary Structure Prediction — MFE Fold (Zuker / Turner)

Fold an RNA to its minimum-free-energy secondary structure in the browser — dot-bracket, MFE, base pairs and an arc diagram, no ViennaRNA install.

🌐 Runs on the SeqBench API — also callable via REST & MCP, and in bulk from the batch tools

Paste an RNA sequence (or DNA — it's treated as RNA) and fold it to its minimum-free-energy secondary structure entirely in your browser. The engine is a from-scratch Zuker dynamic program using the Turner 1999 nearest-neighbor stacking free energies with length-based hairpin/bulge/internal-loop penalties and a linear multiloop model — the same algorithm class as RNAfold/mfold. There's no usable browser build of ViennaRNA (and its license restricts commercial redistribution), so this reimplements the well-established method natively. You get the dot-bracket structure, the estimated MFE, the base-pair list, and an arc diagram you can export. Because the loop energy model is simplified (no full terminal-mismatch/special-hairpin/coaxial-stacking tables and no pseudoknots), treat the MFE as a solid comparative estimate — great for comparing candidates or checking an oligo's self-structure — rather than a lab-grade absolute.

How to use the RNA Structure (MFE) tool

  1. 1Paste an RNA/DNA sequence (up to a few hundred nt) or load the tRNA example.
  2. 2Read the MFE, base-pair count and dot-bracket structure, and inspect the arc diagram (each arc is a base pair).
  3. 3Export the arc diagram, or copy the dot-bracket — and remember it's a simplified-model estimate, best used comparatively.

Frequently asked questions

What algorithm and energies does it use?

A Zuker minimum-free-energy dynamic program (the V/W/WM recursions) with Turner 1999 nearest-neighbor stacking free energies, length-based hairpin/bulge/internal-loop initiation penalties, a Ninio asymmetry term, and a linear multiloop model. It predicts one MFE structure and does not allow pseudoknots — standard for this class of tool.

How accurate is the MFE?

It's a good comparative estimate, not a lab-grade absolute. The loop energy model is deliberately simplified relative to a full ViennaRNA/mfold parameter set (no complete terminal-mismatch, special-hairpin, or coaxial-stacking tables), so absolute kcal/mol values will differ somewhat from those tools. It's well-suited to ranking candidates, checking whether a region tends to be structured, or evaluating an oligo's self-structure.

Why not just use ViennaRNA?

There is no legitimate, maintained WebAssembly build of ViennaRNA that runs in a browser, and ViennaRNA's own license restricts commercial redistribution. This tool is a clean-room native TypeScript implementation of the standard MFE algorithm so it can run fully client-side with no install and no server compute.

Is there a size limit?

Yes — folding is O(n³), so inputs are capped at a few hundred nucleotides to stay responsive in the browser. Fold a specific region rather than a whole transcript.

Is my sequence uploaded, and can I use this from code?

The fold runs on SeqBench's stateless API (the same pure function) and nothing is stored. The rna_fold tool is available via the REST API and MCP server.

Related tools