spatial-domains

spatial-domains

When to use

Find tissue domains from expression and coordinates. Use spatial-annotate for named cell labels and spatial-genes for variable genes.

Use from a step

from skills._sdk.notebook import load_skill, read_input, write_output
library = load_skill("spatial-domains")
data = read_input("input.h5ad")
data = library.identify(data, method="leiden", spatial_weight=0.3)
write_output(library.domain_counts(data), "tables/results.csv")

Run examples/example_step.py through the step runner.

API

identify(adata, *, method: str='leiden', resolution: float=1.0, spatial_weight: float=0.3, refine: bool=False, random_state: int | None=None, **parameters)

Identify domains in place and return the same AnnData.

Reads log-normalized X, X_pca and spatial coordinates; graph methods reuse existing expression neighbors. SpaGCN, STAGATE and BANKSY results vary between runs because their training wrappers do not expose every RNG.

:param adata: Preprocessed spatial AnnData; expression values are retained. :param method: CLI default leiden, or louvain/spagcn/stagate/graphst/banksy/cellcharter. :param resolution: Graph-clustering resolution, CLI default 1.0. :param spatial_weight: Spatial graph weight for Leiden/Louvain, CLI default 0.3. :param refine: False by default; True smooths labels using spatial KNN. :param random_state: None preserves CLI defaults: backend seed 42 for STAGATE, GraphST and CellCharter, otherwise 0; PCA uses 0. An explicit integer overrides both PCA and supported backend seeds. Existing PCA is reused. :param parameters: Backend options listed in references/parameters.md; fixed-K methods use n_domains=7 unless supplied. :returns: The same AnnData with spatial_domain and JSON run diagnostics. :raises ValueError: Unsupported method or invalid graph parameters. :raises ImportError: A backend is missing; use install_skill_deps with the named package.