README_repo.md

Spatial Transcriptomics Pipeline

Analysis pipeline for MERSCOPE spatial transcriptomics data, covering spatial embedding, batch integration, cell type annotation, and downstream analyses.

Repository structure

Folders

Global-analyses/

Core pipeline — runs on every dataset.

Notebook	Description
Manual_slide_cluster.ipynb	Optional Stage 0. Interactive Napari-based tool to draw ROIs on a spatial section and split it into separate .h5ad files. Use when one slide contains multiple distinct tissue sections.
01_per_section_banksy.ipynb	Preprocesses each section (normalize → log1p) and runs BANKSY spatial embedding (λ=0.8). Saves two files per section: *_banksy_genes.h5ad (genes only) and *_banksy_full.h5ad (1500 features including neighbourhood).
02_global_integration_scvi.ipynb	Concatenates all *_banksy_genes.h5ad files, runs QC, trains scVI for batch correction, and clusters with Leiden at multiple resolutions.
02_1_global_integration_scvi.ipynb	Re-integrates a cell subset (e.g. immune cells) with refined scVI parameters. Supports Apple Silicon MPS acceleration.
04_banksy_aware_integration.ipynb	Integrates all *_banksy_full.h5ad files using PCA + Harmony on the full 1500-feature BANKSY matrix. Better for identifying spatially-defined tissue domains.

---

Spatial-neighbourhood/

Identifies tissue domains that recur across conditions.

Notebook	Description
Banksy_domain_analysis.ipynb	Runs BANKSY + Harmony per group (one injury type × timepoint), extracts domain expression profiles, then clusters domains cross-condition into meta-domains via hierarchical correlation clustering.

---

Add-on-analyses/

Cell type annotation and downstream analyses on the integrated object.

Notebook	Description
Tangram.ipynb	Maps a multiome scRNA-seq reference (immune cells + fibroblasts) onto spatial cells using Tangram to transfer fine-grained cell type labels.
Augur.ipynb	Ranks immune cell types by transcriptional perturbation (AUC) across injury models and phases using Augur (via pertpy).
Commot.ipynb	Infers spatial cell-cell communication using COMMOT and the CellChat ligand-receptor database (mouse, 50 µm threshold).

---

Differential-gene-expression/

Pseudobulk DEG analysis between injury models.

Notebook	Description
DEG.ipynb	Pseudobulk differential expression between two injury conditions, stratified by phase. Three methods provided: Wilcoxon (quick exploration), edgeR (recommended), DESeq2 (Python-native alternative). Includes automatic bias correction for transcript detection efficiency differences across slides.

---

Converstion_R_Python/

Utilities for moving data between R and Python.

File	Description
Data_extraction_for_Python_conve…	Exports count matrices and metadata from a Seurat object to files readable by Python/AnnData.
Seurat_object_conversion_part2…	Converts a Seurat object to AnnData format.
Tutorial	Step-by-step guide for the conversion workflow.

---

Requirements

Two conda environments cover the full pipeline:

Environment	Used for
banksy	01_per_section_banksy, Manual_slide_cluster
scvi	All other notebooks

banksy: scanpy · banksy · scipy · scikit-learn · napari · shapely · magicgui

scvi: scvi-tools · scanpy · pytorch · lightning · pertpy · commot · tangram-sc · harmonypy · pydeseq2 (optional)

For DEG with edgeR: R installation with the edgeR Bioconductor package.

Input format

Each input .h5ad must have:

• adata.X — raw count matrix (integers)
• adata.obs["x"], adata.obs["y"] — spatial coordinates
• adata.obs["batch"] or adata.obs["sample"] — slide/section ID