Spatial Pattern Analysis

Overview

This skill answers: Which genes/features show significant spatial organization in this tissue?

Two complementary analyses:

1. Spatially Variable Genes (SVG): Find genes with non-random spatial expression
2. Spatial Statistics: Quantify and test spatial patterns

Decision Tree: Which Analysis?

Q: What do you want to know?
│
├─ Which genes have spatial patterns?
│   └─ SVG Detection
│       ├─ SpatialDE - Gaussian Process, handles noise well
│       ├─ SPARK-X - Fast, scalable to large datasets
│       └─ Moran's I per gene - Classic, interpretable
│
├─ Are spatial patterns statistically significant?
│   └─ Spatial Statistics
│       ├─ Global Moran's I - Overall spatial autocorrelation
│       ├─ Getis-Ord Gi* - Local hot/cold spot detection
│       └─ Ripley's K/L - Point pattern analysis
│
└─ Where are the spatial hotspots?
    └─ Local Spatial Statistics
        ├─ LISA - Local Moran's I
        └─ Getis-Ord Gi* - Hot spot analysis

Spatially Variable Gene Detection

Method Selection

Workflow

Step 1: Run SVG Detection

Use find_spatial_genes tool with:

• method: "spatialDE", "sparkx", or "moran"
• n_top_genes: Number of top genes to return (default 100)

Step 2: Interpret Results

Key output columns:

• pval / padj: Statistical significance
• moranI / effect_size: Strength of spatial pattern
• gene: Gene identifier

Filtering criteria:

Recommended thresholds:
- Adjusted p-value < 0.05 (stringent: < 0.01)
- Effect size > 0.1 (moderate spatial pattern)
- Expression level: Filter very low-expressed genes

Step 3: Visualize Top SVGs

Use visualize_data with plot_type="spatial":

• Show expression of top SVGs spatially
• Overlay with tissue domains for context

Spatial Statistics Analysis

Global Statistics

Moran's I: Measures overall spatial autocorrelation

• Range: -1 (dispersed) to +1 (clustered)
• 0 = random distribution
• Use for: Overall pattern assessment

Geary's C: Alternative to Moran's I

• Range: 0 (clustered) to 2 (dispersed)
• 1 = random
• More sensitive to local variation

Local Statistics

LISA (Local Moran's I): Identifies local clusters

• High-High: Hot spots (high values surrounded by high)
• Low-Low: Cold spots (low values surrounded by low)
• High-Low / Low-High: Spatial outliers

Getis-Ord Gi*: Hot spot analysis

• Positive z-score: Hot spot (high value cluster)
• Negative z-score: Cold spot (low value cluster)
• Use for: Precise hotspot identification

Workflow

Step 1: Compute Spatial Statistics

Use analyze_spatial_statistics tool with:

• statistic: "moran", "geary", "getis_ord", or "ripley"
• feature: Gene or obs column to analyze

Step 2: Interpret Results

Common Analysis Scenarios

Scenario 1: Find Tissue Boundary Genes

Goal: Genes marking transitions between regions
Approach:
1. Run SVG detection (SpatialDE recommended)
2. Overlay top SVGs with domain boundaries
3. Look for genes with gradient patterns

Scenario 2: Identify Hot Spots of Activity

Goal: Where is gene X most active?
Approach:
1. Run Getis-Ord Gi* on gene expression
2. Map significant hot spots spatially
3. Cross-reference with cell type composition

Scenario 3: Compare Spatial Organization

Goal: Do two conditions differ in spatial organization?
Approach:
1. Compute Moran's I for each condition separately
2. Compare I statistics
3. Test for significant difference

Biological Interpretation

What SVGs Tell You

• Boundary markers: Genes defining tissue compartments
• Gradient genes: Continuous spatial variation
• Spot-specific: Localized expression patterns
• Tissue architecture: Structural organization genes

Combining with Other Analyses

Troubleshooting

Output Summary

Successful spatial pattern analysis provides:

• SVG list: Ranked spatially variable genes
• Statistics: Significance and effect sizes
• Spatial maps: Visualization of top patterns
• Hotspots: Localized areas of interest