Mantpy: extracellular-matrix analysis for spatial omics

Mantpy is a scverse-based framework for graph analysis of the extracellular matrix (ECM) in spatial omics. It represents cells and ECM patches as distinct, linked node types, so matrix structure can be analysed on its own or together with cellular context. Mantpy works with AnnData and interoperates with Scanpy, Squidpy, and other single-cell and spatial tools.

Installation

Mantpy requires Python 3.11 or newer.

pip install mantpy

Optional extras add heavier dependencies only when needed:

pip install "mantpy[gnn]"      # graph learning and explainability
pip install "mantpy[patch]"    # learned image-patch features
pip install "mantpy[spatial]"  # SpatialData integration
pip install "mantpy[segment]"  # Cellpose segmentation

Quick start

This runs as written — toy_ecm_roi synthesises a small region of interest in memory, so there is nothing to download:

import mantpy as mt

# A small synthetic ROI: image stack, channel panel, cell table.
roi = mt.datasets.toy_ecm_roi()

# Read multiplexed imaging and a cell table into AnnData.
adata = mt.io.read_imc(
    roi.image, panel=roi.panel, cells=roi.cells,
    sample_id="toy", condition="ctrl",
)

# Normalise channels and segment the ECM into patch nodes.
mt.pp.normalize(adata)
mt.pp.extract_ecm_patches(
    adata,
    roi.image,
    ecm_channel="ColIV",
    ecm_K="auto",
    features=["mean"],
)

# Build cell, ECM, and joint cell-ECM graph layers.
mt.gr.build_graph(adata, mode="cell")
mt.gr.build_graph(adata, mode="ecm")
mt.gr.build_graph(adata, mode="cell_ecm")

# Quantify spatial organisation.
mt.tl.cell_ecm_enrichment(adata, cell_type="B")
mt.tl.neighbourhood_clustering(adata, n_clusters=4)

# Visualise results.
mt.pl.cell_ecm_graph(adata)
mt.pl.neighbourhood_clusters(adata)

Point mt.io.read_imc at your own files to run the same pipeline on real data:

adata = mt.io.read_imc("image.tiff", panel="panel.csv", cells="cells.csv")

The toy ROI is synthetic — its clusters and statistics describe the generator, not biology. Use it to learn the API, then move to real data.

Tutorials

Step-by-step guides, each executed when the documentation is built:

• Loading data
• Building ECM graphs
• Cell–ECM graphs
• Exporting to AnnData
• Exporting to PyTorch Geometric

The full manuscript analyses run on real cohorts in the reproducibility repository, each opening in Colab.

Public datasets

One-line, checksummed loaders for every tutorial dataset, cached outside the package and reused offline:

intestine = mt.datasets.coliv_intestine()   #  93 MB
lung = mt.datasets.balbc_pbs_lung()         #  71 MB
liver = mt.datasets.schistosoma_ecm()       #  38 MB
prostate = mt.datasets.prostate_he_visium() # 194 MB

Every loader returns the same Bunch shape as toy_ecm_roi, so moving an example from synthetic to real data is a one-line change. The bundles are frozen under CC BY 4.0 in an immutable Zenodo record.

Documentation

Documentation · API reference · Changelog

Contact

For questions and help requests, use the scverse discourse. To report a bug, use the issue tracker.