cellxgene-census

Query single-cell expression data at population scale using the Census API to filter by cell type, tissue, disease, and metadata across 61M+ cells in thousands of datasets, retrieving raw expression matrices and standardized annotations.

Yes. Census integrates directly with PyTorch and scanpy workflows, enabling you to retrieve expression matrices, embeddings, and metadata for downstream machine learning and bioinformatic analyses without manual format conversion.

Use Census to query and aggregate expression data across thousands of datasets with TileDB-SOMA filtering, enabling large-scale cross-dataset comparisons, exploratory discovery, and model training on harmonized, population-scale single-cell data.

No. Census enables targeted queries with obs and var filters to retrieve only cells matching your criteria—by cell type, tissue, disease, or custom metadata—avoiding full dataset downloads and supporting out-of-core analysis workflows.

Census provides versioned data access via the census_version parameter, ensuring reproducible analyses by specifying exact snapshots of CZ CELLxGENE data and enabling context-managed resource handling for reliable, large-scale queries.