Spatial

Spatial workflows operate on AnnData objects with coordinates in obsm['spatial'] — the layout used by Visium-class datasets.

Requirements

• Uploaded or converted .h5ad with spatial coordinates
• Typically one spot (or cell) per row in obs
• Gene expression matrix aligned to spot barcodes

The test fixture data/test_visium_brain.h5ad (~1155 spots) is included for local development.

Spatial domains step

Under Analyze → Find Structure → Spatial domains, the pipeline:

1. Builds a spatial neighbor graph
2. Clusters spots into tissue domains (Leiden on spatial + expression features)
3. Identifies spatially variable genes (SVGs) in the engine

Explore spatial viewer

Explore → Spatial renders spots in coordinate space with:

• Domain or cluster coloring
• Gene expression overlay on selection
• Pan and zoom for tissue navigation

H&E histology image pyramids are not yet wired into the UI (the tiling engine writes OME-Zarr pyramids, but the viewer only renders spots).

Spatial results also include neighborhood enrichment, domain-by-condition comparison, and lightweight niche candidates from enriched domain co-localization, shown below the spot viewer.

Reference deconvolution

A standalone Reference Deconvolution Analyze step estimates cell-type composition fractions from sample- or spot-level expression profiles when marker-gene overlap is sufficient.

Enrichment and DE

Use standard DE and enrichment steps on spatial datasets when contrasts or domains define groups of interest. Domain labels appear in obs after a spatial run completes.

Planned extensions

• Tiled H&E image overlay
• Platform-specific ingest (Xenium, MERFISH, CosMX)