This repository contains the reproducibility package for the OmicOS H&E analyses used in Fig. 5 panels c-m and the related supplementary H&E panels. It keeps the lightweight plotting tables, rendered panels, notebooks, provenance manifests, and scripts needed to audit or redraw the released figures.
Large raw whole-slide images, pretrained model weights, and bulky upstream intermediates are not tracked in git. Their required locations and hashes are documented in the manifests and in the external-data README files under analysis/inputs/, analysis/intermediates/, and data/sfig1/he_roi_sources/raw_he/.
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
pytest -qThe original local environment used for generation was /opt/miniforge/envs/omicverse_working. On the source workstation the same checks can be run with:
/opt/miniforge/envs/omicverse_working/bin/python -m pytest tests -qTo rebuild artifacts rather than inspect the committed lightweight bundle, point OMICOS_SOURCE_ROOT at a complete OmicOS workspace containing the upstream analysis directories and raw inputs:
cp env.template.sh env.sh
$EDITOR env.sh
source env.sh
make build-main
make build-sfig1
make build-sfig2The full analysis-level pipeline additionally needs the external raw H&E inputs, pretrained weights, and upstream intermediates:
make run-analysis
make verify-analysisMissing external files fail loudly with FileNotFoundError; the scripts do not silently replace unavailable raw data with placeholders.
OmicOS-HE_Reproducibility/
├── README.md
├── LICENSE
├── pyproject.toml
├── Makefile
├── env.template.sh
├── fig5_main.ipynb
├── fig5_supplementary_1.ipynb
├── fig5_supplementary_2.ipynb
├── data/ lightweight plotting data and thumbnails
├── outputs/ rendered panels and exported tables
├── scripts/ rebuild scripts for main and supplementary figures
├── analysis/ analysis-level provenance package
│ ├── code/ pipeline scripts and source snapshots
│ ├── final_plotting_data/ analysis-local copy of lightweight plotting data
│ ├── outputs/ rendered analysis-local notebooks and panels
│ ├── inputs/README.md raw input policy; heavy files not tracked in git
│ └── intermediates/README.md upstream intermediate policy; heavy files not tracked in git
└── tests/ release and figure contract tests
| Artifact class | Why omitted | Where documented |
|---|---|---|
| Raw Visium HD and TCGA LUAD whole-slide images | Multi-GB files exceed normal GitHub limits and may carry source-provider terms | analysis/MANIFEST.json, analysis/inputs/README.md |
| GigaPath and STPath pretrained weights | Large model artifacts should be fetched from their original providers | analysis/MANIFEST.json, analysis/inputs/README.md |
| Upstream feature arrays and copied OmicOS intermediates | Regeneratable from full local sources; several files exceed 100 MB | analysis/MANIFEST.json, analysis/intermediates/README.md |
| Supplementary Fig. 1 raw H&E BTF | Same raw Visium HD tissue image as the full analysis input | data/sfig1/he_roi_sources/raw_he/README.md |
The lightweight CSV, JSON, PNG, PDF, and notebook outputs required to inspect the published panels remain tracked.
The release contract tests check that repository metadata exists, executable notebook sources use checkout-relative paths, required plotting columns are present, image artifacts needed for plotting exist, and heavy external data policy is explicit.
make testFor the full analysis package on a workstation with the external heavy inputs restored:
make verify-analysis@misc{omicos-he-reproducibility-2026,
title = {OmicOS H&E Reproducibility Package},
author = {OmicVerse contributors},
year = {2026},
url = {https://github.com/omicverse/OmicOS-HE_Reproducibility}
}