cellxgene-census

Querying single-cell genomics data from CELLxGENE Census uses SOMA-based value_filter syntax to narrow results by cell identity, tissue, disease, and donor metadata across obs and var dimensions. You can retrieve specific cell types like CD4 T cells in brain tissues by applying these metadata-driven filters.

Loading large single-cell expression datasets without memory limits requires out-of-core axis queries for large result sets, while smaller queries use in-memory AnnData loading. This scalable data retrieval method handles 61M+ cells by iterating chunked matrices rather than loading everything at once.

SOMA querying syntax filters single-cell RNA-seq metadata by applying explicit value_filter conditions across obs and var dimensions. You must use strict primary-data filtering with is_primary_data set to True and specify explicit census versioning for reproducible queries.

You can use scanpy and PyTorch dataloaders with CELLxGENE Census data through integration-ready workflows. The system provides patterns for scanpy preprocessing and experimental PyTorch training dataloaders to feed extracted AnnData subsets directly into downstream analysis.

You need to pin a specific CELLxGENE Census version for reproducible single-cell analysis to ensure stable dataset access. The system requires explicit census versioning and supports opening stable or pinned releases, preventing data drift when querying 61M+ single cells.