Workflows

A workflow connects inputs, tools, and decision steps into a reusable analysis. A workflow run records one execution and its jobs, results, and credit usage. Use a batch when you only need to repeat one tool across many inputs.

Workflow components

The editor represents an analysis as connected nodes:

• Input: Supplies sequences, structures, ligands, or other supported data.
• Tool: Runs an analysis using its connected inputs and configured settings.
• Connection: Passes compatible output to a later node.
• Decision: Selects which candidates continue, using configured criteria or a human review.
• Merge: Combines compatible data streams for later steps.

One tool node can create multiple jobs. The number depends on the inputs and how that tool processes records and combinations.

Create a workflow

Open Workflows:

1. Select New workflow.
2. Choose a project if you want to organize the workflow.
3. Select Start from scratch for a blank canvas, or Browse templates for an existing analysis pattern.
4. In Editor, select Add block in the bottom toolbar, then choose an input or tool.
5. Select each node to configure it. Connect its right-side output handle to the next tool's left-side input handle.

Select a node to inspect its configuration. Review the data type and format on connections, particularly when a tool produces several kinds of output. A sequence and a structure are different inputs even when they describe the same protein.

Connect formats

Connections use the selected output and destination input, including their scientific type and file format. Equivalent format names, such as phy and phylip, do not require a conversion and leave the file contents unchanged.

When a connection needs a supported conversion, the editor offers a choice of conversion steps. You can also select the destination tool and use Find format conversion. Review the proposed tools and outputs before inserting them. The inserted tools are visible in the workflow and contribute to its job count and credit estimate.

A conversion must preserve the information required by the next analysis. For example, converting a docked ligand to SMILES retains its molecular description but loses its docked coordinates. Creating a new structure from that SMILES does not recover the pose. Select a pose-preserving structure output when continuing docking results into pose assessment or simulation.

Pair inputs by identity

Under Job creation, All combinations creates a job for each permitted combination of connected inputs. Pair by index uses list order. When records have shared identifiers, use Match inputs by metadata fields (optional) with All combinations to pair by those identifiers instead.

Enter one or more field names separated by commas, such as target_id, compound_id. Every field must match. A text ID and a numeric ID are different values. Missing keys stop the step, unmatched records create no jobs, and repeated matching IDs can create multiple jobs. Existing scientific candidate relationships are retained.

The estimate uses known matching IDs when available. If an earlier tool has not produced its IDs yet, the estimate can be higher than the final job count. Review the estimate and workflow spending limits before running.

Combine scores and preserve files

Add a Merge node to combine results from several branches. Choose the behavior that fits the analysis:

Mode	Result
Concatenate	All input records, including duplicates.
Union (Remove duplicates)	The first record for each matching key, or each distinct complete artifact if no key is selected.
Intersection	Records from the first input whose keys occur in every input.
First input only	The first connected input unchanged.
Join score rows (matching IDs)	Combined columns for matching IDs across all inputs.
Join score rows (keep first input)	Every record from the first input, with matching columns from the other inputs.

Score joins require Match by metadata fields. Columns stay in separate groups, such as input_1.score and input_2.score, so one tool's score cannot overwrite another's. Repeated IDs produce every matching combination; missing keys are errors.

CSV cells remain text unless you name their columns under Numeric CSV columns. Blank numeric cells remain missing values, and invalid numbers are rejected. Keep identifiers as text when leading zeros matter.

Choose the joined output according to what comes next. Score rows are suitable for reviewing or downloading a combined table. To continue into a scientific tool, choose the connected input whose original file should travel with the joined scores. Its file contents and identity are retained; the scores become metadata for filtering or ranking.

The selected file source stays attached to its connection when inputs are reordered. If that connection is removed, choose a new source. For a left join, select the first input if unmatched rows must continue; a row without a file in the selected input stops the step.

Select the best candidates within groups

Rank orders candidates and Top K keeps the requested number. Use a numeric metadata field such as input_2.score, then choose whether the highest or lowest value is best.

Set Group by metadata field (optional) to rank separately for each target, compound, or other group. For example, a limit of one with target_id keeps the best candidate for each target. Candidates without a group value are excluded. Numeric ranking does not treat text values as numbers.

Save and revise

Editor changes are saved automatically. Check the save indicator before leaving; Save now saves pending changes and Retry save retries a failed save.

A draft contains your edits. A published revision is the saved definition used for execution, including through the Workflows API. Revision history lets you inspect earlier revisions and restore one into an editable draft. Restoring does not rewrite the definitions used by past runs.

Publishing an executable revision does not make the workflow public.

Run a workflow

Before starting, supply data for every required input and resolve the editor's validation errors:

1. Review tool settings and connections.
2. Open the cost estimate and check the expected number of jobs.
3. Configure any spending limits.
4. Select Run workflow.

Starting from the editor saves pending changes and prepares a published revision for the run. The view switches to Runs after submission.

The run uses a snapshot of its definition and inputs. The graph is locked while an active run prevents editing. Each job is charged when it is ready to run; the full estimate is not charged at startup.

Inspect runs and results

Open Runs and select an execution. The run view shows step status, child jobs, errors, and credit usage.

Select a tool step's results to inspect its jobs and outputs. Use Download all for available run files, or open an individual job for its result views and downloads.

When a run has several attempts, choose the attempt you want to inspect. Keep its identity with exported results so you can distinguish the original execution from a rerun.

Review a paused run

A pause can require different actions:

• Review required: Select Review outputs, inspect the candidates, choose at least one, and use Approve … and continue. The decision becomes part of the run history.
• Insufficient credits or spending limit: Resolve the funding constraint, then select Try again.
• Other checkpoint: Read the pause message and use Continue when its requirements are satisfied.

Continuing keeps the same attempt. Waiting alone does not resolve a checkpoint that requires your decision or a higher credit balance.

Cancel or rerun

Use Cancel on an active run to request cancellation. Review the resulting job states and credit totals; cancellation does not erase earlier output.

A Rerun from… action starts a new attempt within the same run. Eligible earlier outputs can be reused, while newly launched jobs use credits. Review the selected starting step and rerun costs before continuing.

Troubleshooting

Use the message on the affected node or run to locate the problem:

• Missing input: Open the named input node and supply data in a supported format.
• Invalid connection: Check the source output and destination input types, formats, and required slots.
• Run cannot start: Resolve validation, saving, access, or plan-limit errors shown by the editor.
• Paused run: Inspect the pause reason and complete its required action.
• Failed step: Read the job error and any partial outputs before rerunning it.