Goal: run three independent single-chain predictions through a small pipeline, then reconcile every input with its job and output. Complete the single-sequence tutorial first so you know what one valid result looks like.
Inspect the three-entry library
Download ubiquitin-variants.fasta and ubiquitin-run-manifest.csv.
| Identifier | Sequence change, numbered from 1 | Length |
|---|---|---|
ubiquitin_WT |
Reference sequence | 76 |
ubiquitin_I44A |
Isoleucine 44 → alanine | 76 |
ubiquitin_K48R |
Lysine 48 → arginine | 76 |
The two variants are illustrative synthetic inputs, not experimentally characterized examples. A similar predicted fold would not establish preserved function or a beneficial mutation. The manifest starts with empty job and output columns; fill them with real returned values.
Build a single-stage pipeline
Open Pipelines, add a Sequence List input and an ESMFold2 service node, and connect the input to Input Sequences. The connection supplies the sequence input’s supported schema.
- Open the sequence selector and import the FASTA as three named protein entries.
- Confirm all identifiers and lengths. Set one copy of each sequence.
- Configure the service as in the pilot:
single_sequenceMSA mode, one diffusion sample, 20 loops, and 100 sampling steps. Record other defaults. - Inspect the input and service node sizes. The intended size is 3, with one sequence per independent prediction.
- Keep the graph to these two nodes for the first bulk run. No downstream file selector is needed yet.
Checkpoint: this is three monomer predictions, not one three-chain complex. A normal service form can combine several sequence entities into a single complex; use the pipeline’s list expansion and review its validated size. If the builder does not show three independent service inputs, stop and inspect the configuration.
Review and submit
Expect three service executions, each with one requested diffusion sample. The earlier pilot is a separate job. Use the pipeline’s estimate; do not assume that the total is exactly three times a previous charge.
Resolve validation errors and give the pipeline a recognizable descriptor such as Tutorial — three ubiquitin variants. Export the template for your records, then submit once. Keep the pipeline ID. Templates omit private sequence/file data, so retain the FASTA separately and reattach it after import.
Reconcile each input
Open the pipeline status page and inspect every child item. Fill in the downloaded CSV using actual IDs and output names:
| Manifest field | What to record |
|---|---|
candidate_id / sequence |
Preserve the supplied identity and exact sequence |
pipeline_id |
The parent workflow ID |
job_id |
The child prediction job for that input |
status |
Actual pending, running, completed, or failure state |
structure_file |
Returned structure filename for that child job |
review_notes |
Confidence, molecular contents, and issues found |
Do not infer correspondence from download order or filenames alone. Verify each child job’s inputs against the manifest. If one item fails, retain its row and error, inspect the cause, and reconcile existing jobs before retrying it. Do not rerun the entire pipeline merely to replace one failed item.
Compare and extend deliberately
Inspect each structure and its confidence using the same procedure as the pilot. Compare matched residues with consistent alignment and atom selections. Record uncertainty as well as apparent differences; predictions alone do not measure stability or activity.
Only after all three inputs are accounted for should you add another stage. Run one upstream example, inspect actual filenames and formats, then configure a Job File Select node to select exactly the intended sample. A larger sample count or a broad filename expression can multiply downstream work.
The builder broadcasts size-one inputs and pairs equal-size lists by index. For example, one receptor and three ligands give three pairs; three receptors and three ligands do not automatically give nine combinations. See Pipelines and Bulk submissions before expanding the experiment.