What problem does it solve?
Spatial transcriptomics data combines gene expression with tissue coordinates, but turning raw Visium, MERFISH, or Slide-seq output into biological insight requires a long chain of QC, normalization, spatial clustering, deconvolution, and ligand-receptor analysis that is easy to get wrong. This Skill provides a structured eight-phase workflow that guides an AI agent through the entire analysis, from data import to a publication-style report.
Core Features & Use Cases
- Spatial domain identification: Graph-based Leiden clustering with spatial constraints (squidpy) segments tissue into coherent expression domains with marker genes.
- Spatially variable gene detection: Moran's I autocorrelation testing (FDR < 0.05) classifies genes into gradient, hotspot, boundary, and periodic patterns.
- Cell type deconvolution and communication: Integrates scRNA-seq references via Cell2location, Tangram, or SPOTlight, and maps ligand-receptor interactions using OmniPath and squidpy ligrec.
- Dataset discovery: HuBMAP, OmicsDI, GEO, and CELLxGENE tools locate published spatial datasets for reference or validation.
- Use Case: Given a 10x Visium breast cancer section, the workflow identifies tumor core, invasive margin, and immune infiltrate domains, maps T cell exclusion, and highlights PD-L1/PD-1 checkpoint hotspots at the tumor-immune interface.
Quick Start
Analyze my 10x Visium dataset in the spaceranger output folder to identify spatial domains and spatially variable genes, then generate a full tissue report.
