pyXenium 0.4.2

Xenium I/O, multimodal analysis, topology workflows, contour-native spatial profiling, GMI inference, mechanostress analysis, and optional external workflow bridges.

pyXenium

nine feature areas for Xenium I/O, multimodal analysis, CCI, pathway topology, contour geometry, GMI inference, mechanostress analysis, and external workflow bridges.

PyPI · Read the Docs · GitHub · Changelog · Releases

Figure 1. pyXenium summarizes Xenium I/O, multimodal analysis, spatial topology, contour-aware inference, mechanostress analysis, and external workflow bridges in one submission-ready overview.

pyXenium is a Python toolkit for 10x Genomics Xenium organized around nine major sections.

Section	Canonical entry	Start here
Xenium I/O	pyXenium.io	Xenium data loading guide
Multimodal Analysis	pyXenium.multimodal	Multimodal overview
Cell-Cell Interaction	pyXenium.cci	CCI tutorial hub
Pathway Topology	pyXenium.pathway	Pathway tutorial
Contour Geometry	pyXenium.contour	Contour tutorial hub
GMI Inference	pyXenium.gmi	Contour GMI guide
Mechanostress	pyXenium.mechanostress	Mechanostress tutorial
AI-Driven Spatial Pathologist	pyXenium.spatho	RTD bridge guide
SpatialPerturb Bridge	pyXenium.perturb	SpatialPerturb bridge guide

Legacy compatibility entry points under pyXenium.analysis, pyXenium.validation, and pyXenium.io.load_xenium_gene_protein(...) remain importable, but new code should target the canonical pyXenium namespaces above. The spatho and SpatialPerturb workflows are installed and run separately; pyXenium does not vendor them or add them as core runtime dependencies.

Release & Build

• Current repository version: 0.4.2
• Package index: PyPI
• Documentation site: pyxenium.readthedocs.io
• Canonical build status: GitHub Actions CI
• Supported Python: >=3.8
• License: pyXenium Non-Commercial License

Install

pip install pyXenium==0.4.2

For local development:

git clone https://github.com/hutaobo/pyXenium
cd pyXenium
pip install -e ".[dev]"

For documentation work:

pip install -e ".[docs]"

For the optional SpatialPerturb Bridge runtime on Python 3.9+:

pip install -e ".[perturb]"

Representative examples

Xenium I/O

from pyXenium.io import read_xenium

slide = read_xenium("/path/to/xenium_export", as_="slide", prefer="zarr")

XeniumSlide is the canonical in-memory object. The old XeniumSData name remains as a compatibility alias, and as_="sdata" still works for legacy code.

To build Atera-style learning stores with contour-segmented H&E crops:

pyxenium slide build-atera --atera-root Y:\long\10X_datasets\Xenium\Atera --output-root D:\GitHub\stGPT\outputs\xenium_slides\atera

This writes one xenium_slide.zarr per case plus slide_manifest.json, qc_report.json, cell_to_contour.parquet, structure_assignments.csv, and contour_patches_manifest.json. Raw Xenium outs directories are read only; per-cell H&E patches are not generated.

Canonical multimodal loading

from pyXenium.multimodal import load_rna_protein_anndata

adata = load_rna_protein_anndata(
    base_path="/path/to/xenium_export",
    prefer="auto",
)

Contour expansion

from pyXenium.contour import expand_contours

expand_contours(
    sdata,
    contour_key="protein_cluster_contours",
    distance=25.0,
    mode="voronoi",
)

Contour-GMI inference

from pyXenium.gmi import ContourGmiConfig, run_atera_breast_contour_gmi

result = run_atera_breast_contour_gmi(
    dataset_root="/path/to/WTA_Preview_FFPE_Breast_Cancer_outs",
    output_dir="pyxenium_gmi_outputs/atera_s1_s5",
    config=ContourGmiConfig(feature_count=500, spatial_feature_count=100),
)

pyXenium.gmi is a canonical beta surface: the API is public, while biological interpretation should be checked with the bundled controls, cross-validation, and PDC Slurm reproducibility workflow. The Atera S1/S5 validation completed on PDC Dardel in the v0.4.1 release, supporting an S5/DCIS RNA program led by NIBAN1 and SORL1 under the QC20 primary model.

Mechanostress analysis

from pyXenium.io import read_xenium
from pyXenium.mechanostress import MechanostressConfig, run_mechanostress_workflow

sdata = read_xenium("/path/to/xenium_export", as_="sdata", include_boundaries=True)
result = run_mechanostress_workflow(
    sdata,
    output_dir="pyxenium_mechanostress_outputs/sample_1",
    config=MechanostressConfig(),
)

For cohorts organized as one Xenium sample per directory, use:

from pyXenium.mechanostress import MechanostressConfig, run_mechanostress_cohort

cohort = run_mechanostress_cohort(
    "/path/to/headandneckSCC",
    output_dir="pyxenium_mechanostress_outputs/hnscc",
    config=MechanostressConfig(coupling_genes=("KRT5", "EPCAM")),
)

pyXenium.mechanostress is a canonical beta surface for extracting mechanical stress signals from Xenium morphology and spatial cell context.

AI-Driven Spatial Pathologist via spatho

AI-Driven Spatial Pathologist is an external workflow layer for AI-assisted pathology review around Xenium-scale spatial transcriptomics. Its public Python package and CLI are named spatho.

pip install -U spatho
spatho init-workflow --organ breast --case-name breast_case_01 --dataset-root /path/to/Xenium_outs --output workflow.json
spatho doctor --config workflow.json
spatho run --config workflow.json

In pyXenium, this is documented as an optional external workflow bridge rather than a new pyXenium.spatho namespace. The handoff is possible because XeniumSData keeps the cell table, transcript points, cell/nucleus boundaries, H&E image metadata, and SpatialData-compatible organization together for downstream tools.

SpatialPerturb Bridge via SpatialPerturb

SpatialPerturb is an external workflow package for combining spatial transcriptomics with Perturb-seq references. pyXenium exposes a lightweight pyXenium.perturb bridge that writes a handoff JSON and stable external CLI commands without vendoring the SpatialPerturb algorithms.

from pyXenium.perturb import SpatialPerturbBridgeConfig, write_spatialperturb_handoff

spec = write_spatialperturb_handoff(
    SpatialPerturbBridgeConfig(
        xenium_path="/path/to/Xenium_outs",
        output_dir="spatialperturb_reports/breast_case_01",
        cell_group_path="/path/to/cell_groups.csv",
        roi_geojson_path="/path/to/xenium_explorer_annotations.geojson",
        sample_name="breast_case_01",
    ),
    "spatialperturb_bridge.json",
)
print(spec["command_text"]["run_reference_benchmark"])

SpatialPerturb Bridge scores mean Perturb-seq-derived program similarity projected onto Xenium tissue. They do not mean the tissue cell contains the corresponding knockout, guide, or drug perturbation.

Documentation entry points

The docs now separate the nine feature sections from the main reading paths:

• Quickstart
• Tutorials hub
• User guide
• Workflows
• API reference
• Changelog

Start here: pyxenium.readthedocs.io

License

Copyright (c) 2025-2026 Taobo Hu. All rights reserved.

This project is source-available, not open source. You may use, modify, and redistribute it only for non-commercial purposes under the terms of the LICENSE file. Commercial use requires prior written permission from the copyright holder.