Goal: generate a single-chain ubiquitin prediction, inspect its confidence, and compare it with an experimental reference. You will submit one ESMFold2 job and retain its structures, confidence files, and job ID.
Prepare the example
Download ubiquitin.fasta and the 1UBQ experimental reference. The sequence is the 76-residue chain A in RCSB PDB entry 1UBQ. The experimental coordinates are for comparison after inference; do not upload them as a prediction input.
>ubiquitin_WT
MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGG
Check that you have one protein entity, 76 residues, one copy. This example has no ligand, additional chain, modification, or custom alignment. See the input provenance for source URLs and checksums.
Configure one prediction
Open ESMFold2 and select Input Sequences. Upload the FASTA or add one protein entry named ubiquitin_WT and paste only the sequence into its sequence field.
Use these settings for a bounded exploratory run; check the live form if a service update changes the available controls:
| Field | Tutorial setting | Reason |
|---|---|---|
| MSA Mode | single_sequence |
Run without an alignment search |
| Custom MSA | Empty | No custom alignment in this example |
| Diffusion Samples | 1 |
Request one structure rather than an ensemble |
| Number Loops | 20 |
Keep the current default refinement count |
| Sampling Steps | 100 |
Keep the current default sampling count |
| Other advanced settings | Current defaults; record them | Make the configuration reproducible |
Add the note Tutorial — ubiquitin_WT — single sequence. Review the estimated credits, then submit once. Save the job ID. Follow its state in Workspace; on a network interruption, reconcile job history before retrying.
Inspect the output
When the job completes, open the result page and its file listing. Download the ranked structure in the format actually returned (ESMFold2 returns mmCIF structures) and the associated confidence files.
- Open the structure in the result viewer. Confirm a single protein chain and the expected sequence length.
- Inspect global fold confidence and per-residue confidence, using the definitions reported by the tool. Locate lower-confidence segments in the structure.
- Check that confidence values and file identifiers correspond to this sample. With one monomer, interface scores may be absent or inapplicable.
- Save the structure and the raw confidence files alongside the job ID and settings.
Checkpoint: you should have one prediction job and its sample’s structure and confidence outputs. A completed job without the expected molecular contents needs investigation before downstream use.
Compare with the experimental reference
Open 1UBQ.pdb in a structure viewer. Align the predicted protein to experimental chain A using a structural alignment tool with explicit sequence correspondence. Confirm residue numbering and matched atoms before interpreting an RMSD.
Compare the compact fold, the placement of secondary structure, and the C-terminal tail. Examine local differences alongside confidence and experimental context; an isolated prediction and a crystal structure need not have identical conformations.
If you calculate a quantitative comparison, record the matched residues, atom selection, alignment method, and units. The SDK’s alignment and RMSD reference documents its matching rules. It requires matching backbone atoms and residue identities; resolve chain names, numbering, or missing atoms before using it.
This familiar protein is useful for learning the workflow. Agreement with one known structure does not establish performance on unfamiliar targets, and high confidence does not demonstrate biological activity.
Record the outcome and continue
Keep a short record with ubiquitin_WT, job ID, settings, structure filename, confidence summary, and observations from the reference comparison. No exact confidence or RMSD is prescribed here: report what your run produced.
Continue to the three-sequence bulk tutorial to repeat the process with explicit input identities, or read Structure prediction before choosing a method for your own system.