Fix ESMFold failures — non-standard residues like B, J, O and X, the 10-2,500 limit, timeouts on long chains, low pLDDT, and a missing PAE artifact.

When ESMFold fails to submit or returns a structure that looks wrong, this page covers the common failure modes — from sequence validation and length limits to long runtimes, low pLDDT confidence, and missing PAE outputs.

Common issues and fixes

ESMFold uses the ESM-2 language model to predict structures directly from sequences. Understanding common issues can help you get better predictions and interpret results appropriately.

Sequence validation failures

Ensure only the 20 standard amino acids are present; remove B, J, O, U, X, or Z.
Remove whitespace, numbers, and FASTA comments inside the sequence body.
Check that sequence length is within the supported limit (10-2,500 residues).

Long runtimes or timeouts

Split very long sequences into domains if biological context allows.
Reduce batch size to one sequence per job for long proteins.
Verify cluster capacity and retry during lower-traffic windows if queues are long.

Low pLDDT or high PAE

Flag low-confidence regions (<50 pLDDT) as disordered instead of forcing rigid models.
Consider trimming long unstructured tails and re-running to improve core confidence.
Use external secondary structure or disorder predictors to corroborate low-confidence zones.
Remember that ESMFold's pLDDT scores are calibrated to correlate with structural accuracy; consistently low scores often indicate genuinely disordered regions rather than model failure.

Novel or designed sequences

Because ESMFold does not use MSA, it can predict structures for sequences without homologs. However:

Designed proteins or de novo sequences may have lower confidence if they differ significantly from natural protein distributions
Synthetic peptides with non-natural features may produce unreliable predictions
For engineered proteins, focus on high-pLDDT regions and validate experimentally when possible

Missing PAE artifacts

Ensure returnPaE is enabled; it is true by default.
If outputs are still missing, re-run with a smaller sequence to rule out memory constraints.

Still stuck?

Provide the job ID, FASTA header, sequence length, and any custom parameters (maxTokens, seed, returnPaE).
Share whether the issue is validation, runtime, or confidence-related to speed up support.

ESMFold troubleshooting