---

name: external-integrations description: "Use Scanpy optional integrations, scanpy.external wrappers, Dask-backed arrays, and GPU/RAPIDS handoffs without over-installing optional dependencies." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

Scanpy External Integrations

Use this sub-skill when a task involves optional dependency extras, scanpy.external, Dask-backed arrays, RAPIDS/GPU acceleration, external exporting, or integration/imputation/demultiplexing/trajectory wrappers outside Scanpy's core preprocessing and tool APIs.

Route By Need

• Optional extras planning: Read references/optional-dependencies.md before recommending installs; choose the narrow extra or upstream package for the requested feature and do not suggest installing every extra.
• External wrapper selection: Read references/external-api-map.md to map scanpy.external.pp, scanpy.external.tl, scanpy.external.pl, and scanpy.external.exporting methods to prerequisites, outputs, and moved APIs.
• Dask or large arrays: Read references/dask-and-rapids.md before promising distributed support; Scanpy supports selected Dask-backed preprocessing paths, not every algorithm or external package.
• GPU acceleration: Route GPU requests to references/dask-and-rapids.md; GPU workflows are usually delegated to rapids-singlecell, while Scanpy's RAPIDS neighbor backend requires a compatible cuML stack.
• Import/install failures: Use references/troubleshooting.md plus scripts/check_scanpy_optional_deps.py to diagnose missing optional packages without installing or importing heavy dependencies.

Safe Helper

Run the bundled checker from this sub-skill directory when you need a machine-readable optional-dependency inventory:

python scripts/check_scanpy_optional_deps.py --feature scanorama --feature magic --json
python scripts/check_scanpy_optional_deps.py --list-features

The helper uses importlib.util.find_spec and package metadata only. It has no network, install, import-heavy, or package-mutation side effects.

Boundaries

• Prefer core Scanpy guidance for ordinary normalization, PCA, neighbors, clustering, plotting, and AnnData handling unless the task explicitly needs optional integrations, Dask-backed arrays, or GPU handoffs.
• Prefer sc.pp.harmony_integrate, sc.pp.scrublet, and sc.pp.scrublet_simulate_doublets for moved APIs; legacy scanpy.external.pp aliases are deprecated compatibility routes.
• Keep install advice minimal: scanpy[leiden], scanpy[skmisc], scanpy[plotting], scanpy[magic], scanpy[scanorama], scanpy[bbknn], scanpy[dask], scanpy[dask-ml], and scanpy[scrublet] are targeted choices for specific features.
• For wrappers without Scanpy extras, recommend the named upstream package only for that method, such as mnnpy, phate, palantir, trimap, sc-sam, phenograph, harmonyTS, wishbone, pypairs, or cellbrowser.
• Do not recommend broad optional extras, development/test dependency groups, source checkout scripts, or original repository paths for runtime use.