ChatSpatial: Skill for Claude Code

Differential Analysis

Overview

This skill answers: What genes/features differ between groups in this spatial data?

Three main comparison types:

1. Marker genes: What defines each cluster/cell type?
2. Condition comparison: How does treatment/disease change expression?
3. Regional comparison: What differs between spatial domains?

Decision Tree: Which Comparison?

Q: What are you comparing?
│
├─ Clusters or cell types
│   └─ Marker Finding
│       ├─ Wilcoxon rank-sum - Default, robust
│       ├─ t-test - Parametric, fast
│       └─ Logistic regression - Handles confounders
│
├─ Experimental conditions (treated vs control)
│   └─ Condition Comparison
│       ├─ Pseudobulk + DESeq2 - Gold standard for replicates
│       ├─ MAST - Single-cell aware
│       └─ Mixed models - Complex designs
│
└─ Spatial domains or regions
    └─ Regional Comparison
        ├─ Same as marker finding
        └─ Consider spatial autocorrelation

Marker Gene Finding

Purpose

Identify genes that distinguish one group from all others (1 vs rest) or from a specific group (pairwise).

Method Selection

Method	When to Use	Strengths
Wilcoxon	Default choice	Non-parametric, robust
t-test	Large datasets	Fast, well-understood
Logreg	Confounders present	Adjusts for covariates

Workflow

Step 1: Define Groups

Ensure grouping column exists in adata.obs:

• Clusters: leiden, louvain
• Cell types: cell_type, annotation
• Domains: spatial_domain

Step 2: Run Marker Finding

Use find_markers tool with:

• groupby: Column name defining groups
• method: "wilcoxon", "t-test", or "logreg"
• n_genes: Number of top markers to return per group

Step 3: Interpret Results

Key output columns:

• names: Gene name
• scores: Test statistic
• pvals_adj: Adjusted p-value (FDR corrected)
• logfoldchanges: Log2 fold change
• pct_nz_group: Percent expressing in group
• pct_nz_reference: Percent expressing in other groups