Skip to content

Latest commit

 

History

History
216 lines (160 loc) · 12 KB

public_datasets.md

File metadata and controls

216 lines (160 loc) · 12 KB

Public Datasets for Development & Testing

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.


What ships with TissuePlex, and how

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.md rather 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.

Classic Visium — reader verification set

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_scalef is 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_fullres is 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; deriving pixel_size from 55 µm is wrong by 18%.
  • Registration: 98.1% of in_tissue spots 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|Lrp2 and Cst3|Lrp2 — megalin-mediated proximal tubule uptake, which is the expected kidney biology.

Tier 1 — Tiny Datasets (10–30 MB) — Primary Development

Artificially cropped to three 640px patches across two FOVs. Full Xenium output folder structure intact. Ideal for rapid iteration and CI testing.

Xenium v3.0.0 Tiny

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

Xenium v4.0.0 Tiny

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.


Tier 2 — Small Full Datasets (278–379 MB) — Performance & Realism Testing

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

Tier 3 — Full Public Datasets (GBs) — Stress Testing & Showcase

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

Other Public Sources

Zenodo

NCBI GEO


Recommended Download Script

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 -lh

Dataset Feature Coverage Matrix

Use 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

Notes on Xenium Output Format Versions

The Xenium output folder structure has evolved across software versions. Key differences that may affect parsing:

  • v2.x: transcripts.csv.gz + transcripts.parquet both present; cells.zarr.zip for 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.


Reference