MALDI-MSI Analysis — Mouse Urinary Bladder

This repository demonstrates a complete MALDI-MSI data analysis pipeline on a public benchmark dataset — from raw data loading to statistical interpretation of tissue regions. Originally developed as teaching material for private tutoring in mass spectrometry imaging analysis, following the preprocessing approach described in Bemis, Föll et al. (2023, Nature Methods).

---

Dataset

Mouse urinary bladder — Römpp et al. (2010) Angew. Chem. Int. Ed. 49:3834
PRIDE: PXD001283
Instrument: AP-SMALDI LTQ Orbitrap, 10 µm spatial resolution
Format: Centroided imzML — 34,840 pixels × ~153 peaks after preprocessing

---

Analysis pipeline

Notebook 01 — Data Exploration and Preprocessing

• What is MALDI-MSI data and how it is structured
• TIC map and mean spectrum
• Pixel-level comparison (centre vs periphery)
• Four-step preprocessing: 10 ppm rebinning → TIC normalisation → 2.5% frequency filtering → rebinning to reference peaks
• Step-by-step preprocessing visualisation with diagnostic figures
• Top peaks per lipid class (LPC, PE, PC, TAG)
• PCA before and after preprocessing (scree plot + spatial colouring)
• t-SNE before and after preprocessing

Notebook 02 — Tissue Segmentation

• Choosing k: elbow curve + silhouette score + biological evidence
• K-means, GMM (tied covariance), Spectral Clustering (n_neighbors=50)
• Pairwise ARI comparison with Hungarian label alignment
• Agreement map — where all three methods consistently agree
• Pairwise disagreement maps
• Difference spectrum at tissue boundaries
• PCA and t-SNE coloured by cluster assignments

Notebook 03 — Statistical Analysis

• Pairwise Mann-Whitney U tests (not one-vs-rest) per region pair
• Benjamini-Hochberg FDR correction
• Marker deduplication (100 ppm minimum spacing)
• Volcano plots — one per region vs most contrasting partner
• Top 3 marker ion images per region
• Mean spectra per region with peak annotation
• Global colocalization analysis (top 15 peaks, Pearson correlation)

---

How to run

1. Open in GitHub Codespaces
2. Install dependencies:

pip install pyimzml scikit-learn seaborn matplotlib pandas scipy numpy -q

3. Download dataset (see data/README.md) — ~815 MB
4. Run notebooks in order: 01 → 02 → 03
   • Notebook 02 loads results saved by Notebook 01
   • Notebook 03 loads results saved by Notebook 02

---

Repository structure

src/
  load.py           — imzML loading and image reconstruction
  preprocessing.py  — 4-step preprocessing pipeline
  visualisation.py  — plotting functions and assign_lipid_class
  stats.py          — fdr_correction, differential_abundance, colocalization

notebooks/
  01_data_exploration_preprocessing.ipynb
  02_segmentation.ipynb
  03_statistical_analysis.ipynb

results/
  figures/          — all generated figures (tracked by git)
  tables/           — CSV output tables (tracked by git)

---

References

1. Römpp et al. (2010) Angew. Chem. Int. Ed. 49:3834 — dataset
2. Bemis, Föll et al. (2023) Nature Methods 20:1883 — preprocessing pipeline
3. Deininger et al. (2011) Anal. Chem. 84:1277 — TIC normalisation
4. Alexandrov et al. (2010) J. Proteome Res. 9:6535 — K-means for MSI
5. Benjamini & Hochberg (1995) J. R. Stat. Soc. B 57:289 — FDR correction
6. Mann & Whitney (1947) Ann. Math. Stat. 18:50 — Mann-Whitney U test
7. Hubert & Arabie (1985) J. Classification 2:193 — ARI
8. Von Luxburg (2007) Stat. Comput. 17:395 — spectral clustering
9. Hsu & Turk (2009) J. Chromatogr. B 877:2714 — lipid identification by mass

---

Author

Reza Rajaee