Spatial transcriptomics
Analyze Visium-class spatial datasets — cluster tissue domains, identify spatially variable genes, measure neighborhood enrichment, compare domains across conditions, nominate co-localization niche candidates, and explore gene expression in tissue coordinate space.
Research question
What spatial domains exist in this tissue section, which domains co-localize, which genes vary across space, and where are genes of interest expressed relative to tissue architecture?
Who this is for
- Spatial biology labs working with Visium or similar spot-based platforms
- Tumor microenvironment researchers mapping expression to tissue regions
- Neuroscience groups analyzing spatially resolved brain sections
Data requirements
| Data | Required | Purpose |
|---|---|---|
.h5ad with spatial coordinates | Yes | Spots in obsm['spatial'] |
| Gene expression per spot | Yes | Domain clustering and overlays |
| H&E image pyramid | No | Planned — not yet wired into UI |
Local development fixture: data/test_visium_brain.h5ad (~1155 spots).
Workflow
Upload spatial .h5ad → Spatial domains → Neighborhood enrichment → Explore spot viewer → DE by domain → Enrichment
Step 1 — Upload spatial dataset
Create a study with Spatial modality (or upload a spatial .h5ad into an existing study). Confirm coordinates exist in obsm['spatial'] and barcodes align with obs.
Step 2 — Spatial domain clustering
Under Analyze → Find Structure → Spatial domains, the pipeline:
- Builds a spatial neighbor graph
- Clusters spots into tissue domains (Leiden on spatial + expression features)
- Identifies spatially variable genes in the engine
- Tests which domain pairs neighbor each other more often than expected by label permutation
Domain labels appear in obs after the run completes. The spatial result payload also includes SVG rankings, neighborhood enrichment, domain-by-condition summaries when condition metadata exists, and lightweight niche candidates from enriched domain co-localization.
Step 3 — Explore spatial viewer
Open Explore → Spatial to render spots in coordinate space:
- Color by domain or cluster assignment
- Select genes for expression overlay
- Review enriched neighboring domain pairs below the viewer
- Pan and zoom across the tissue section
H&E histology image pyramids remain a planned extension. Reference deconvolution is available as a standalone Analyze step for expression profiles or spots with sufficient marker-gene overlap.
Step 4 — Differential expression and enrichment
Run standard DE and enrichment steps using domain labels or saved contrasts as grouping variables. Compare domains of interest or condition labels when present in metadata.
Step 5 — Snapshot
Create a snapshot in History to preserve the spatial analysis parameter set and completed run IDs.
Expected outputs
- Spatial domain assignments per spot
- Spatially variable gene rankings and overlays
- Neighborhood enrichment and co-localization table for spatial domains
- Domain-by-condition abundance comparison
- Lightweight spatial niche candidates
- Reference deconvolution fractions and fit diagnostics when the standalone deconvolution step is run
- Interactive spot viewer with domain and gene coloring
- DE and enrichment results stratified by domain
- Run history and snapshots for reproducible review
Integration with other areas
- Pathology provides whole-slide imaging and tissue segmentation for morphology-guided spatial context
- Oncology adds TME-specific interpretation and clinical endpoint integration on top of spatial compbio pipelines
Compbio owns spatial omics ingestion and analysis; adjacent areas add domain-specific overlays without duplicating the spatial engine.