cellxgene-census

To query single-cell transcriptomics metadata without downloading everything, you can extract versioned cell metadata, dataset summaries, and gene coverage slices directly from the CZ CELLxGENE Census using TileDB-SOMA-compatible query syntax.

You retrieve AnnData slices for specific tissues, diseases, or cell types by applying primary-data filtering and targeted expression queries to the Census, yielding objects ready for downstream analysis in Scanpy.

Yes, you can prepare machine learning datasets at scale by using the Census out-of-core matrix iteration and presence matrices, which support memory-safe analysis and large-scale workflows without loading entire datasets into memory.

Yes, reproducible analysis requires correct Census version selection, primary-data filtering, and TileDB-SOMA-compatible query syntax to ensure your versioned single-cell and spatial transcriptomics queries remain consistent across runs.

The best way to export spatial data slices is to use the Census spatial data export workflows, which apply primary-data filtering to extract targeted expression matrices and metadata into AnnData formats for spatial transcriptomics tasks.

Out-of-core iteration handles population-scale exploration by processing large single-cell matrices in chunks, allowing you to inspect cell metadata and gene coverage across organisms without exhausting memory resources.