Skip to content

Adding a new pipeline

CBIcall is designed to be extensible: you can add new analysis pipelines without changing the core framework as long as the workflow wiring is declared in the workflow registry.

In CBIcall, the source of truth for available workflows is:

  • workflows/config/cbicall.workflows.yaml (registry; what exists)
  • workflows/schema/workflows.schema.json (schema; what a valid registry looks like)

The Python code validates user parameters (enums, defaults, compatibility rules) and then resolves the correct workflow script from the YAML registry.

1. How CBIcall discovers workflows

At runtime, CBIcall:

  1. Reads the user parameter YAML (e.g. params.yaml)
  2. Applies defaults and validates semantic rules (pipeline/mode/genome/engine compatibility)
  3. Loads and schema-validates the workflow registry:
  4. workflows/config/cbicall.workflows.yaml
  5. workflows/schema/workflows.schema.json
  6. Resolves workflow scripts from the registry and checks:
  7. referenced files exist
  8. Bash scripts are executable (+x)
  9. Dispatches the workflow through the selected engine (Bash or Snakemake)

Key point: adding a pipeline is mostly YAML + scripts, not Python.

2. Create workflow scripts

Create one or more workflow entrypoints under the engine directory and GATK version:

Bash

workflows/bash/gatk-4.6/
  mypipe_single.sh
  mypipe_cohort.sh
  parameters.sh

Snakemake

workflows/snakemake/gatk-4.6/
  mypipe_single.smk
  mypipe_cohort.smk
  config.yaml

Naming conventions

CBIcall follows the pattern:

  • {pipeline}_{mode}.{sh|smk}
  • pipeline: e.g. wes, wgs, mit, mypipe
  • mode: single or cohort

3. Register the pipeline in the workflow registry

Edit workflows/config/cbicall.workflows.yaml and add your pipeline under the appropriate engine + version.

Minimal example (Bash, single mode)

workflows:
  bash:
    base_dir: "workflows/bash"
    versions:
      gatk-4.6:
        common:
          parameters: "parameters.sh"
          coverage: "coverage.sh"
          jaccard: "jaccard.sh"
          vcf2sex: "vcf2sex.sh"
        pipelines:
          mypipe:
            single: "mypipe_single.sh"

Adding cohort mode

        pipelines:
          mypipe:
            single: "mypipe_single.sh"
            cohort: "mypipe_cohort.sh"

Snakemake example

workflows:
  bash:
    base_dir: "workflows/bash"
    versions:
      gatk-4.6:
        common:
          parameters: "parameters.sh"
          coverage: "coverage.sh"
          jaccard: "jaccard.sh"
          vcf2sex: "vcf2sex.sh"
        pipelines:
          wes:
            single: "wes_single.sh"

  snakemake:
    base_dir: "workflows/snakemake"
    versions:
      gatk-4.6:
        common:
          config: "config.yaml"
        pipelines:
          mypipe:
            single: "mypipe_single.smk"

The schema currently requires workflows.bash to exist, even if you primarily use Snakemake. Keep a minimal bash block if needed.

4. Expose user parameters via the parameter YAML

Users select the workflow by providing:

pipeline: mypipe
mode: single
workflow_engine: bash
gatk_version: gatk-4.6
projectdir: cbicall

Then add pipeline-specific fields as needed (reference, targets, etc.). The recommended approach is:

  • keep “policy” options in YAML (things the user should control)
  • keep script-specific internal wiring inside the workflow scripts

5. Single vs cohort mode

If you support both modes, register both in the YAML and provide both entrypoints.

If cohort mode does not make sense, omit it from the registry. CBIcall will raise a clear error if a user requests an unavailable mode.

6. Validation and guardrails

CBIcall will fail fast when:

  • pipeline, mode, workflow_engine, gatk_version are invalid
  • the pipeline/mode combination is not allowed for the selected GATK version
  • incompatible combinations are selected (example: snakemake with gatk-3.5 if restricted)
  • a registry entry points to a missing file
  • a Bash workflow script is not executable (+x)

This keeps “what is allowed” stable and “what is wired” configurable.

Create a tiny, deterministic example to support CI and user onboarding:

examples/mypipe_test/
  params.yaml
  input/
  expected/

Goals:

  • fast execution (ideally < 1 minute)
  • deterministic outputs
  • validates one or two key artifacts

8. Document the pipeline page

Add a new page:

docs/pipelines/mypipe.md

Include:

  • requirements (tools, references)
  • supported modes
  • required YAML parameters
  • example command(s)
  • output layout

Then add it to mkdocs.yml:

nav:
  - Pipelines:
      - Adding a pipeline: pipelines/adding-a-pipeline.md
      - MyPipe: pipelines/mypipe.md

9. Optional: Container support

If the pipeline requires extra software, pin versions for reproducibility:

  • extend your Dockerfile, or
  • publish a new container image

Make sure the workflow scripts remain runnable in the container environment.

Quick checklist

  • Add workflow script(s) under workflows/{bash|snakemake}/{gatk-version}/
  • Register pipeline in workflows/config/cbicall.workflows.yaml
  • Ensure registry passes workflows/schema/workflows.schema.json
  • Make Bash scripts executable (chmod +x)
  • Add example params.yaml
  • Add docs page and link it in mkdocs.yml