CZ CELLxGENE Discover Census
Chan Zuckerberg Initiative
Query tens of millions of single cells from hundreds of studies as if they were one table.
The Census packages every single-cell RNA data set in CZ CELLxGENE Discover into one harmonized, queryable store with standardized cell type, tissue, disease, and assay labels. You slice it by metadata in Python or R and get AnnData or Seurat objects back, without downloading each study. The 2025 long-term release added rhesus macaque, marmoset, and chimpanzee to human and mouse, and spatial assays are in testing.
Access
- Cost
- FreeFree via Python and R packages reading from public AWS storage.
- License
- Open licenseData released under CC BY 4.0 per CELLxGENE Discover
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadTileDB-SOMA, AnnData (h5ad), Seurat
What's in it
- One row is
- One row per cell, with a gene-by-cell count matrix
- Covers
- Not geographic; studies worldwide
- Years
- Studies published roughly 2015-present
- Updated
- Irregular
- Size
- Tens of millions of cells; long-term-support builds plus weekly builds
Quality
Standardized metadata ontologies make cross-study queries practical. Batch effects between studies remain and labels are only as good as each author's annotation.
- Strong batch effects across studies
- Cell type labels come from submitting authors
- Duplicate cells across overlapping data sets need filtering
Connects to
- Allen Brain Cell Atlas: Compare brain cell types with a single-lab whole-brain atlas
- GTEx Portal: Bulk tissue expression versus cell-type-resolved expression
- Human Protein Atlas: Its single-cell section draws on similar public data sets
Questions it could answer
- In which cell types is a drug target gene expressed across the whole body?
- How do immune cell proportions differ between healthy and diseased lung?