spatial-omics

Spatial Omics Analysis

Method selection and execution guidance for spatial transcriptomics and spatial proteomics technologies.

Technology Detection

Identify the spatial platform first — resolution and data structure dictate the analysis approach:

Technology?
├── Visium (10x Genomics)
│   ├── Resolution: ~55 um spots, each covering ~1-10 cells
│   ├── Data: spot x gene count matrix + tissue image + spot coordinates
│   ├── Coordinate type: grid (hexagonal array)
│   └── Load: sq.read.visium() or sd.read_10x_visium()
│
├── MERFISH / seqFISH / Xenium (single-molecule FISH)
│   ├── Resolution: subcellular, single-molecule
│   ├── Data: molecule coordinates → cell x gene matrix after segmentation
│   ├── Coordinate type: generic (continuous coordinates)
│   └── Load: sd.read_xenium() or custom from segmentation output
│
├── Slide-seq / HDST (bead-based capture)
│   ├── Resolution: ~10 um beads (near single-cell)
│   ├── Data: bead x gene count matrix + bead coordinates
│   ├── Coordinate type: generic
│   └── Load: custom AnnData with .obsm["spatial"]
│
└── CODEX / MIBI / IMC (spatial proteomics)
    ├── Resolution: single-cell (after segmentation)
    ├── Data: cell x protein intensity matrix + coordinates
    ├── Coordinate type: generic
    └── Load: custom AnnData with .obsm["spatial"]

Analysis Decision Tree

Spatial Neighbors Graph

Building spatial graph (foundation for all spatial stats):
├── Visium → sq.gr.spatial_neighbors(adata, coord_type="grid")
│   Uses hexagonal grid adjacency, not distance
├── All other technologies → sq.gr.spatial_neighbors(adata, coord_type="generic")
│   ├── n_neighs=6 (default, good starting point)
│   └── Or radius-based: radius=float for distance threshold
└── Result stored in adata.obsp["spatial_connectivities"], adata.obsp["spatial_distances"]

Spatial Domain Identification

Approach?
├── Graph-based clustering
│   ├── Standard → Leiden on spatial graph (sq.gr.spatial_neighbors → sc.tl.leiden)
│   └── Combined expression + spatial → compute joint graph (expression kNN + spatial kNN)
└── Visium-specific
    └── BayesSpace (R via rpy2, Bayesian spatial clustering, respects tissue morphology)