Demo data¶

Every tutorial on this site runs on the same public datasets, published on Zenodo under DOI 10.5281/zenodo.22549610 (CC-BY-4.0) — this DOI always resolves to the latest version, so it's the link to share. The commands below fetch v1.2 (Zenodo record 22549611), which is version-pinned so the MD5s below always match.

What's in the box¶

Three single-cell AnnData (.h5ad) files — unpaired scRNA-seq and scATAC-seq from COVID-19 and healthy PBMC donors, plus their pre-computed scGLUE-integrated object.

File	Contents	Shape	Units	Size	MD5
`test_RNA.h5ad`	scRNA-seq counts	29,989 cells × 14,881 genes	8 samples (CoV / HD donors)	94 MB	`69f00904fb9ff21e1d8e83afd63aa441`
`test_ATAC.h5ad`	scATAC-seq counts	29,021 cells × 230,356 peaks	8 samples (SRA accessions)	260 MB	`9053f14b396ae2553677d00b69ba18a8`
`test_multiomics_integrated.h5ad`	scGLUE-integrated RNA + ATAC	59,010 cells × 13,820 genes	29,989 RNA + 29,021 ATAC cells	54 MB	`3d474460e651fe61cc0751a3911512d1`

The third file is optional — the multi-omics tutorial starts from it to skip scGLUE training. It already carries the GLUE joint embedding — the sample-preserved view (obsm['X_glue'], aliased to obsm['Z_rmd']) and the sample-removed view (obsm['Z_comp']) — plus joint cell-type labels.

No metadata CSV needed

All per-cell metadata is already inside each file's .obs: the sample identifier (sample), a severity phenotype (sev.level; 1 = healthy, 4 = severe COVID-19), and — for RNA — a batch column (Aruna vs Wilk) and pre-computed celltype labels. You do not need a separate sample_meta.csv or HVG list.

The RNA and ATAC modalities are unpaired (they share no donors), which is exactly the setting the multi-omics pipeline is built for: GLUE anchors RNA and ATAC into a joint embedding without any cross-modality cell pairing.

Download¶

Grab the files into a local data/ folder:

wgetcurlPython

mkdir -p data
wget -O data/test_RNA.h5ad  "https://zenodo.org/records/22549611/files/test_RNA.h5ad?download=1"
wget -O data/test_ATAC.h5ad "https://zenodo.org/records/22549611/files/test_ATAC.h5ad?download=1"
# optional — used by the multi-omics tutorial:
wget -O data/test_multiomics_integrated.h5ad "https://zenodo.org/records/22549611/files/test_multiomics_integrated.h5ad?download=1"

mkdir -p data
curl -L -o data/test_RNA.h5ad  "https://zenodo.org/records/22549611/files/test_RNA.h5ad?download=1"
curl -L -o data/test_ATAC.h5ad "https://zenodo.org/records/22549611/files/test_ATAC.h5ad?download=1"
# optional — used by the multi-omics tutorial:
curl -L -o data/test_multiomics_integrated.h5ad "https://zenodo.org/records/22549611/files/test_multiomics_integrated.h5ad?download=1"

import urllib.request, os
os.makedirs("data", exist_ok=True)
base = "https://zenodo.org/records/22549611/files"
files = ["test_RNA.h5ad", "test_ATAC.h5ad", "test_multiomics_integrated.h5ad"]
for f in files:
    urllib.request.urlretrieve(f"{base}/{f}?download=1", f"data/{f}")

Verify your download (optional)¶

A corrupted or partial download can surface later as a cryptic HDF5 error (e.g. "bad symbol table node signature"). To catch it early, check the MD5s against the table above:

LinuxmacOS

md5sum data/*.h5ad

md5 data/*.h5ad

Expected (tool-independent <file>: <hash> pairs — md5sum prepends the data/ path and prints hash file, while macOS md5 prints MD5 (file) = hash, so match on the hash regardless of layout):

test_RNA.h5ad:                     69f00904fb9ff21e1d8e83afd63aa441
test_ATAC.h5ad:                    9053f14b396ae2553677d00b69ba18a8
test_multiomics_integrated.h5ad:   3d474460e651fe61cc0751a3911512d1

Citation¶

Jiang, H., & Ji, H. SampleDisco tutorial demo data: scRNA-seq and scATAC-seq COVID-19 PBMC subsets. Zenodo. https://doi.org/10.5281/zenodo.22549610