ESMFold

ESMFold is a no-install online single-sequence structure prediction tool you can run through tools.ranomics.com on a dedicated GPU. The one to reach for on a sequence you designed yourself: it reads the chain directly instead of hunting for natural relatives, which is exactly what a de novo binder does not have.

Fold inputs

Preset

Standalone, one FASTA Paste a single-chain monomer (10 to 400 aa). One fold, inline result. ~30 s on A100-40GB once the 3B model is warm. Batch for many monomers Fold many monomer sequences in one job, up to 500 records. Per-design results stream into the job page as they complete. Cost scales linearly with batch size.

FASTA (single chain)

Paste one monomer sequence in FASTA format (single chain only, 10 to 400 aa). Canonical 20 amino acids plus X for unknown. ESMFold v1 is monomer-only. For multimers use ColabFold (D3) or AF2 standalone (D2). No MSA is computed; the fold comes straight from the ESM-2 language model.

Sequences (FASTA or one per line)

Up to 500 monomer sequences per submit, 10 to 400 aa each. Canonical 20 amino acids plus X for unknown. Lines beginning with > are treated as FASTA headers; lines without a header are auto-named design_0, design_1, … Detected records: 0.

Campaign

Group related runs on your jobs page. Leave blank to file under Uncategorized.

Estimated cost $0.00 based on parameters and historical compute time

Billed by actual GPU consumed, up to the estimate. Surplus returns to your wallet automatically. Cancellation does not refund GPU time already used.

About ESMFold

What it is

Predicts the structure of one protein chain from its sequence alone, in about 30 seconds. It reads the sequence through a protein language model instead of searching for relatives, so it still works on designed or orphan sequences that have no natural family to align against. One chain only — it cannot fold a complex. ESMFold, Lin et al., Science 2023.

When to use

You want a single-chain fold in well under a minute.
Your sequence is designed, or has no known relatives, so a search for them would come back empty anyway.
You are triaging a large batch of sequences and need throughput more than precision.

Prerequisites

Single FASTA sequence (monomer only).
Sequence length under ~600 residues for best accuracy.

Input glossary

Sequence. Single-chain FASTA. Multimers and non-canonical residues are not supported. Use ColabFold or AF2 instead.

Runtime

Preset	Typical min
standalone	~30 s

Predicted PDB with per-residue pLDDT. No PAE (single-sequence prediction has no inter-domain signal). Use as a fast self-consistency check on designed sequences.

References

A run we actually did

The target

Human myelin basic protein — one chain, 304 residues, pasted as plain FASTA. A published reference sequence, not a design.

Myelin basic protein is one of the textbook intrinsically disordered proteins: on its own in solution it has no single fixed structure to predict. That makes it a useful example precisely because the answer is known in advance: it shows what something every user runs into sooner or later actually looks like — a low score coming back, and no obvious way to tell whether the tool failed or the protein did.

What we put in the form

Preset — Standalone, one FASTA: One chain, one fold. The other preset, batch, folds many sequences in a single submission.
FASTA (single chain) — 304 aa, one chain: Pasted as FASTA. ESMFold reads the sequence alone — there is no MSA step and nothing else to configure, which is why it comes back in well under a minute.

What came back

A complete structure, a mean pLDDT of 39.00 and a pTM of 0.119. Open Per-residue pLDDT under the result and the 304-residue strip is red almost end to end: 278 of 304 residues sit below 50, 26 are amber above it, and nothing reaches green or blue anywhere. Not one residue reaches 70; the highest in the chain is 65.9. The longest unbroken stretch that even reaches 50 is 10 residues.

Fold result standalone

mean pLDDT

39.00

pTM

0.119

length

304 aa

Model: facebook/esmfold_v1 · MSA: none · Chains: 1 (monomer-only)

Per-residue pLDDT (304 residues)

blue > 90 · green > 70 · amber > 50 · red ≤ 50

Download PDB

GPU time: 32 seconds.

How to read that

That is the correct answer, not a failed run. The model is not confused about this protein; it is reporting that the protein has no one shape to report, which is what the disorder literature says about it too. A run that genuinely breaks returns an error, not a structure. What separates the two readings is the shape of the per-residue strip, not the mean. Here it is flat and low from end to end — no folded region anywhere. A low mean with a high plateau in the middle and red only at the edges would be the opposite finding: a real folded region with floppy ends, where the middle is trustworthy and only the termini are not. Similar mean, different result, and the mean alone cannot tell them apart. Open the strip.

What happened next

Nothing, because there is nothing here to fix — no setting on this form would raise that number, and running it again returns the same structure. When a chain comes back like this the useful move is to change the question rather than the parameters: fold the fragment that is known to fold, or fold it together with the partner it binds, since many disordered proteins take a definite shape only in complex. If instead you are looking at a designed sequence scoring like this, that is a real failure and the design needs redoing.

Recorded from that one run, which cost $0.04, and took 32 seconds. That is what it did, not a quote for yours — the estimate on the form is the live number for your own inputs.

What good looks like

Each tool reports a subset of these depending on whether it does design, sequence recovery, or structure prediction.

ipTM: Predicted confidence in the contact between two chains, on a 0 to 1 scale. Higher is better: 0.65 to 0.75 depending on the tool. Individual tools set their own pass bar a little either side of that. On a multi-chain target the number may cover the target's own chain–chain interface as well as the binder's.
pLDDT: Per-residue confidence in the predicted fold. Higher means the model is more sure of that part of the structure.
i_pAE and pAE: Predicted alignment error, at the interface (i_pAE) or across the whole structure (pAE). Lower is better.
ProteinMPNN recovery: Fraction of native residues recovered when ProteinMPNN redesigns a known sequence on its native backbone — the folded shape on its own, before a sequence is chosen for it. Higher is better: > 0.4 usable; > 0.6 excellent.

Related tools on Ranomics

If you are picking between ESMFold and a sibling algorithm, these run on the same hub against the same target.

Run AlphaFold2 online: You have a sequence and want the most trusted 3D prediction of it, with per-residue confidence you can act on. It searches for related natural sequences first, which is where the accuracy comes from and where the time goes. For a faster answer use ColabFold or ESMFold; to score a binder against its target use Boltz-2.
Run ColabFold online: You have a sequence and want its 3D shape in a minute or two, trading a little accuracy for speed. It skips the search for related natural sequences that full AlphaFold2 runs. Use it to triage a batch; use AlphaFold2 when the answer has to be right.
Run ProteinMPNN online: You have a backbone — a 3D shape with no sequence decided yet — and need amino-acid sequences that will fold into it. Ranked candidates come back in about 30 seconds. To generate the backbone in the first place, run a binder design tool and feed its PDB in here.