This document lists publicly available datasets suitable for developing and testing TissuePlex across platforms, tissue types, species, and data configurations. It opens with what the repository actually distributes and under which licence; the rest is a catalogue of further public data, mostly 10x Xenium.
| Dataset | Platform | Where it comes from | Licence | Size |
|---|---|---|---|---|
mouse_ileum_tiny |
Xenium | in git — 10x "Xenium Prime Mouse Ileum tiny" crop; edges are synthetic (make_edges.py) because the crop's expression matrix is a near-empty placeholder |
CC BY 4.0 (10x) | 1.6 MB |
visium_hd_tiny |
Visium HD | in git — subset of 10x "Visium HD Tiny Mouse Brain"; see its PROVENANCE.md |
CC BY 4.0 (10x) | 26 MB |
visium_tiny |
Visium | in git — synthetic, make_visium.py |
— | 0.9 MB |
seqfish_synthetic |
seqFISH | in git — synthetic, make_seqfish.py |
— | 0.8 MB |
xenium_human_breast_2fov |
Xenium | release asset via sample_data/fetch_samples.py — trimmed copy of 10x "Human Breast 2 FOV" plus the NICHESv2 run edges/niches_rad30.parquet, 10x's own graph clusters as cell-metadata/, and per-pair summaries as edge-metadata/. Nothing generated. |
CC BY 4.0 (10x) | 274 MB |
visium_mouse_kidney |
Visium | release asset — 10x V1_Mouse_Kidney (two files, see below) plus NICHESv2 edges |
CC BY 4.0 (10x) | 39 MB |
cosmx-mousebrain |
CosMx | release asset — modified subset of the Bruker/NanoString 1000-plex mouse brain as archived by the OSTA project (OSF 5n4q3, folder CosMx1k_MouseBrain2); tx_file.csv truncated to FOV 1 |
Bruker "Data License Agreement for Non-Commercial Use" — PDF ships in the folder; non-commercial only, separate from TissuePlex's own licence | 58 MB |
merscope-vpt-smallset |
MERSCOPE | not distributed — Vizgen's U2OS "small set" from the vizgen-postprocessing vignette (1 GB zip; the cell_by_gene / cell_metadata tables are VPT outputs). No licence is stated for the data and Vizgen's terms of use reserve redistribution, so it stays a local-only development set |
unclear | 43 MB local |
seqfish_instrument2 |
seqFISH | not distributed — Spatial Genomics' seqfish-2-test-dataset.zip used by scverse CI; downloads from spatialgenomics.com are gated and "all rights reserved" |
not open | 2 MB local |
The release assets are attached to the GitHub release tagged v1.0.1 (the code release; the data
zips are extra files on it, pinned by RELEASE_TAG in the script). Each zip holds
only the files TissuePlex reads, a PROVENANCE.md stating source, licence and every
modification, and (for CosMx) the licence PDF. fetch_samples.py verifies a SHA-256
per asset before unpacking.
Two consequences for anyone cutting a new release of the demo data:
- Keep the three licences separate. The 10x sets are CC BY 4.0 and only need
attribution. The Bruker set may be redistributed only with its licence text, a
notice of what was modified, a link to Bruker, and for non-commercial use — none of
which the repository's own licence (AGPL-3.0) can grant, which is why it is a release asset
with its own
PROVENANCE.mdrather than a file in git. - MERSCOPE has no demo a stranger can download. The honest options are a synthetic
fixture in the style of
make_seqfish.py/make_visium.py, or written permission from Vizgen for the small set. Until one of those exists the MERSCOPE reader is exercised only on the authors' machines.
The two datasets the Visium reader was validated on. Both are Space Ranger 1.1.0, which
means they carry the headerless tissue_positions_list.csv — and that is the common
case, because every classic Visium dataset 10x serves without registration predates the
Space Ranger 2.0 rename. The CytAssist-era 2.x downloads return HTTP 403.
You only need two files per dataset, not the full outs/:
B=https://cf.10xgenomics.com/samples/spatial-exp/1.1.0/V1_Mouse_Kidney
mkdir -p sample_data/visium_mouse_kidney && cd sample_data/visium_mouse_kidney
curl -LO $B/V1_Mouse_Kidney_filtered_feature_bc_matrix.h5
curl -LO $B/V1_Mouse_Kidney_spatial.tar.gz
mv V1_Mouse_Kidney_filtered_feature_bc_matrix.h5 filtered_feature_bc_matrix.h5
tar xzf V1_Mouse_Kidney_spatial.tar.gz && rm V1_Mouse_Kidney_spatial.tar.gz| Dataset | Size | Spots in tissue | Notes |
|---|---|---|---|
V1_Mouse_Kidney |
12 MB + 8 MB | 1,438 of 4,992 | The verification dataset. visium_mouse_*/ is gitignored; fetch_samples.py delivers it with its edge layer. |
V1_Adult_Mouse_Brain |
21 MB + 9 MB | — | Used only to cross-check the lattice geometry. |
What they established, none of which the synthetic fixture could:
tissue_hires_scalefis 0.177 and 0.170 — nowhere near the 1.0 that hides a missing coordinate multiply.- The in-row lattice pitch is exactly 138.00 fullres px on both, while
spot_diameter_fullresis 89.45 — a ratio of 0.648, proving that field measures a ~64.8 µm detected footprint rather than the 55 µm capture spot. See the Visium section in CLAUDE.md; derivingpixel_sizefrom 55 µm is wrong by 18%. - Registration: 98.1% of
in_tissuespots land on stained tissue and 99.5% of out-of-tissue spots land on bare slide, measured as mean inverse intensity under each spot polygon in the hires PNG. - NICHESv2 end to end on the kidney: 1,438 spots, 9,666 edges, 2,479 LRMs, 3.2M rows. Top
mechanisms include
Apoe|Lrp2andCst3|Lrp2— megalin-mediated proximal tubule uptake, which is the expected kidney biology.
Artificially cropped to three 640px patches across two FOVs. Full Xenium output folder structure intact. Ideal for rapid iteration and CI testing.
| Dataset | Size | Download |
|---|---|---|
| Mouse Ileum (MultiCellSeg) | 11.9 MB | https://cf.10xgenomics.com/samples/xenium/3.0.0/Xenium_Prime_MultiCellSeg_Mouse_Ileum_tiny/Xenium_Prime_MultiCellSeg_Mouse_Ileum_tiny_outs.zip |
| Mouse Ileum (Nuclear expansion) | 12 MB | https://cf.10xgenomics.com/samples/xenium/3.0.0/Xenium_Prime_Mouse_Ileum_tiny/Xenium_Prime_Mouse_Ileum_tiny_outs.zip |
| Dataset | Size | Download |
|---|---|---|
| Human Kidney (Protein panel) | 20 MB | https://cf.10xgenomics.com/samples/xenium/4.0.0/Xenium_V1_Protein_Human_Kidney_tiny/Xenium_V1_Protein_Human_Kidney_tiny_outs.zip |
| Human Ovary (Nuclear expansion) | 27 MB | https://cf.10xgenomics.com/samples/xenium/4.0.0/Xenium_V1_Human_Ovary_tiny/Xenium_V1_Human_Ovary_tiny_outs.zip |
| Human Ovary (MultiCellSeg) | 29 MB | https://cf.10xgenomics.com/samples/xenium/4.0.0/Xenium_V1_MultiCellSeg_Human_Ovary_tiny/Xenium_V1_MultiCellSeg_Human_Ovary_tiny_outs.zip |
Why use multiple tiny datasets: v3 vs v4 format differences may affect parsing; MultiCellSeg vs nuclear expansion tests different segmentation overlay rendering; mouse vs human tests species-agnostic coordinate handling.
Two-FOV subsets. Realistic transcript density and cell counts. Good for validating rendering performance before testing on full lab data.
| Dataset | Size | Download |
|---|---|---|
| Human Breast Cancer (2 FOV) | 379 MB | https://cf.10xgenomics.com/samples/xenium/2.0.0/Xenium_V1_human_Breast_2fov/Xenium_V1_human_Breast_2fov_outs.zip |
| Human Lung (2 FOV) | 278 MB | https://cf.10xgenomics.com/samples/xenium/2.0.0/Xenium_V1_human_Lung_2fov/Xenium_V1_human_Lung_2fov_outs.zip |
Full-scale datasets for final performance validation and demo. Download selectively — these are large.
Available via the 10x Genomics datasets portal: https://www.10xgenomics.com/datasets
Notable options:
- Human Breast Cancer (full, multiple replicates) — most widely used Xenium reference dataset
- Human Lung Cancer — with multimodal segmentation
- Human Renal Cell Carcinoma (FFPE, Protein + Gene Expression) — tests protein panel rendering
- Human Ovarian Cancer (Xenium Prime 5K) — high gene panel count
- Human Lymph Node (Xenium Prime 5K preview)
- Human Pancreatic Cancer
- Mouse Brain — important for testing non-human tissue and neuronal morphology
- Spatial transcriptomics (Xenium) from early postnatal lung: https://zenodo.org/records/17155546 (~10–11 GB, 5 ROIs)
- Xenium benchmarking data (AnnData format): https://zenodo.org/records/11120307
- Imaging spatial transcriptomics platform benchmarking: https://zenodo.org/records/16848917
- Human brain tissue in situ sequencing: https://zenodo.org/records/15425563
- Retrosplenial cortex single-cell spatial: https://zenodo.org/records/16697913
- GSE264334: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE264334
- GSE300007 — multi-platform imaging spatial transcriptomics comparison (includes Xenium raw data)
Run from the repo root. Downloads Tier 1 (all tiny datasets) and the two Tier 2 datasets.
#!/bin/bash
mkdir -p sample_data
cd sample_data
# Tier 1 — tiny datasets
curl -O https://cf.10xgenomics.com/samples/xenium/3.0.0/Xenium_Prime_MultiCellSeg_Mouse_Ileum_tiny/Xenium_Prime_MultiCellSeg_Mouse_Ileum_tiny_outs.zip
curl -O https://cf.10xgenomics.com/samples/xenium/3.0.0/Xenium_Prime_Mouse_Ileum_tiny/Xenium_Prime_Mouse_Ileum_tiny_outs.zip
curl -O https://cf.10xgenomics.com/samples/xenium/4.0.0/Xenium_V1_Protein_Human_Kidney_tiny/Xenium_V1_Protein_Human_Kidney_tiny_outs.zip
curl -O https://cf.10xgenomics.com/samples/xenium/4.0.0/Xenium_V1_Human_Ovary_tiny/Xenium_V1_Human_Ovary_tiny_outs.zip
curl -O https://cf.10xgenomics.com/samples/xenium/4.0.0/Xenium_V1_MultiCellSeg_Human_Ovary_tiny/Xenium_V1_MultiCellSeg_Human_Ovary_tiny_outs.zip
# Tier 2 — 2-FOV performance datasets
curl -O https://cf.10xgenomics.com/samples/xenium/2.0.0/Xenium_V1_human_Breast_2fov/Xenium_V1_human_Breast_2fov_outs.zip
curl -O https://cf.10xgenomics.com/samples/xenium/2.0.0/Xenium_V1_human_Lung_2fov/Xenium_V1_human_Lung_2fov_outs.zip
# Unzip all
for f in *.zip; do
dirname="${f%.zip}"
unzip "$f" -d "$dirname" && rm "$f"
done
echo "Done. Datasets in sample_data/:"
ls -lhUse this to select datasets that exercise specific viewer features during development.
| Feature to test | Recommended dataset |
|---|---|
| Basic tile rendering + pan/zoom | Mouse Ileum tiny (smallest, fastest) |
| MultiCellSeg segmentation overlay | Mouse Ileum MultiCellSeg or Ovary MultiCellSeg tiny |
| Protein panel / immunostain layers | Human Kidney tiny (protein panel) |
| Human tissue + dense transcripts | Human Breast 2-FOV |
| Large morphology image performance | Human Lung 2-FOV |
| High gene panel count | Human Ovarian Cancer (Xenium Prime 5K) — Tier 3 |
| Non-human / neuronal tissue | Mouse Brain — Tier 3 |
| v3 format compatibility | Any v3.0.0 dataset |
| v4 format compatibility | Any v4.0.0 dataset |
The Xenium output folder structure has evolved across software versions. Key differences that may affect parsing:
- v2.x:
transcripts.csv.gz+transcripts.parquetboth present;cells.zarr.zipfor cell metadata - v3.x: Introduces Xenium Prime panel support; MultiCellSeg segmentation option added
- v4.x: Latest format; protein panel support; updated cell segmentation outputs
The backend Xenium reader should detect the version from experiment.xenium and handle format differences gracefully. Testing against both v3 and v4 tiny datasets during development is strongly recommended.
- 10x Genomics Xenium example data index: https://www.10xgenomics.com/support/software/xenium-onboard-analysis/latest/resources/xenium-example-data
- Xenium output file format documentation: https://cf.10xgenomics.com/supp/xenium/xenium_documentation.html
- 10x Genomics public datasets portal: https://www.10xgenomics.com/datasets