Dataset: Xenium Human Lung Preview — Non-diseased FFPE
Source & Download: 10x Genomics — Xenium Human Lung Preview (standard)
Licensed under CC BY 4.0

• Overview
• Why this detailed pipeline matters
• Repository Contents
• Getting Started
• Figures
• Attribution

This repository implements a transparent and flexible pipeline for processing Xenium spatial transcriptomics data from raw output to spatial visualization.

Unlike workflows that depend on Zarr or aggregated formats, this approach uses step-by-step raw data processing for reproducibility and educational clarity.

1. Reproducibility & clarity
   You manually stream and filter transcripts (e.g., Q ≥ 20) and build the cell×gene matrix, making every step clear and auditable.

2. Robustness to changes
   Xenium data format may change over time; this pipeline handles schema differences gracefully (e.g. variations in column names like x_location vs x_centroid).

3. Scalability & memory efficiency
   The two-pass Arrow/Parquet batching avoids memory issues, allowing for large datasets to be handled smoothly.

4. Customizable & extensible
   Users can easily adjust quality thresholds or extend the pipeline to other spatial platforms (e.g., CosMx, MERFISH).

• scripts/unpack_all.py — Extracts Xenium data from zipped output bundles.
• notebooks/preview_quickstart.ipynb — Walk-through notebook to:
  • Load raw cells.parquet and transcripts.parquet, plus image and metrics files
  • Filter and build a sparse count matrix
  • Construct and QC an AnnData object
  • Normalize, cluster, and visualize spatial patterns
  • Compute neighborhood enrichment and save results

1. Clone this repository

   git clone https://github.com/jrs-orellana/xenium2anndata-analysis-workflow

2. Download the Xenium dataset ZIP from the link above and place it in data/.
3. Run python scripts/unpack_all.py to extract the dataset.
4. Install dependencies (see below).
5. Open and run the notebooks to process the data:
   • 01_xenium_raw2anndata.ipynb — conversion from raw Parquet → AnnData
   • 02_xenium_downstream.ipynb — QC, clustering, marker detection, spatial plots
6. Explore results in results/figures/ and the processed .h5ad file.

Main packages required (see full requirements.txt for exact versions):

• scanpy — single-cell analysis and visualization
• squidpy — spatial transcriptomics analysis
• spatialdata-io — Xenium/Visium data import
• pyarrow — efficient batch processing of Parquet files
• numpy and pandas — data handling
• matplotlib and seaborn — visualization

Install via:

pip install -r requirements.txt

Total Counts	Genes per Cell	Counts vs Genes

Post-QC Density	PCA Scree	UMAP (Leiden)

UMAP: n Genes by Cell	UMAP: Total Counts	Spatial Leiden

Cell-Type Scores	Compartments (High)

Marker Dotplot	Marker Heatmap	Neighborhood Enrichment

If you use this repository or adapt parts of the workflow, please cite it as:

APA style:

Orellana-Montes, J. (2025). xenium2anndata-analysis-workflow: Transparent pipeline for Xenium spatial transcriptomics. GitHub. Available at: https://github.com/jrs-orellana/xenium2anndata-analysis-workflow

BibTeX:

@misc{xenium2anndata2025,
  author       = {Julio Orellana-Montes},
  title        = {xenium2anndata-analysis-workflow: Transparent pipeline for Xenium spatial transcriptomics},
  year         = {2025},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://github.com/jrs-orellana/xenium2anndata-analysis-workflow}}
}

Dataset: Xenium Human Lung Preview — Non-diseased FFPE, 10x Genomics (licensed under CC BY 4.0).
Please cite per 10x Genomics citation guidelines.

• Name: xenium2anndata-analysis-workflow
• Purpose: Detailed, manual parsing and processing of Xenium raw data
• Strengths: Transparency, flexibility, reproducibility over convenience

This project is released under the MIT License. See LICENSE for details.

Dataset belongs to 10x Genomics and is licensed under CC BY 4.0.

Author: Julio Orellana-Montes
For questions, suggestions, or collaborations: open an issue or pull request on GitHub,
or contact me at [email protected]