Pipelines connect input nodes, service nodes, and output-selection nodes into an executable workflow. They are useful for repeatable multi-step calculations and bounded bulk runs. Use the API for logic that needs custom branching, analysis, or external orchestration beyond the builder.
Build the graph
Open Pipelines. Add nodes using the Add control, or start from a pre-made template. A valid graph needs input data and at least one service node.
- Add an input node appropriate for the service field: sequence, structure, molecule, file, or another supported type.
- Supply the input entities and verify their identities and order.
- Add a service node and configure its scientific settings.
- Connect compatible ports. A connection represents a specific field handoff, not merely a conceptual relationship between tools.
- Review validation errors and node sizes before running anything.
Map outputs explicitly
Use a Job File Select node to choose matching job files for downstream steps. Set its filename expression and maximum selected count deliberately. For example, .*\.pdb$ selects a PDB extension; the actual filename must match the upstream service’s outputs.
The extension helps determine compatible downstream fields. A regex that matches nothing, too many files, or an unsupported representation can break the handoff. Run one upstream example and inspect its file listing before defining a selector for an entire campaign.
Do not send a receptor-only structure where the next step expects a full complex, or a relaxed candidate without verifying that identifiers and cofactors were preserved.
Understand node size and pairing
A list node has a size corresponding to its entries. At a service node, connected parent lists must have matching sizes or size one. A size-one input broadcasts; equal-length lists pair by index.
For example, one receptor plus twelve ligand entries produces twelve receptor-ligand inputs. Twelve receptors and twelve ligands pair by index; they do not automatically produce 144 combinations. See Bulk submissions.
Validate and submit
Resolve the builder’s errors, review estimated total credits, and give the pipeline a recognizable descriptor. Submitted pipelines cannot be edited in place. Export a template before running if you need a reusable configuration.
Template codes omit uploaded files and protein data. After importing a code, reattach inputs and validate again; sharing a template is not sharing an executable copy of every private file.
Monitor and recover
Open the pipeline’s status page or find it in the workspace’s Pipelines view. Inspect the jobs and recorded errors for the failed node and item. Keep both pipeline ID and child job IDs.
A successful earlier stage does not prove that every later stage ran. Before rerunning, identify which jobs already exist and whether the failure came from preparation, file mapping, execution, or scientific incompatibility. Avoid resubmitting a complete expensive campaign solely to recover one missing output.