| name | alterlab-cellxgene |
| description | Query the CZ CELLxGENE Census (61M+ cells) programmatically via cellxgene-census and TileDB-SOMA, slicing expression by tissue, disease, or cell type and returning AnnData. Use when pulling reference single-cell RNA-seq data from the largest curated public atlas, running population-scale queries, or benchmarking your data against a reference — for analyzing your own dataset use scanpy or scvi-tools. Part of the AlterLab Academic Skills suite. |
| license | MIT |
| allowed-tools | Read Write Edit Bash(python:*) Bash(uv:*) |
| compatibility | Self-contained — runs under `uv run python` with the skill's Python package installed; no API key or account required. |
| metadata | {"skill-author":"AlterLab","version":"1.0.0"} |
CZ CELLxGENE Census
Overview
The CZ CELLxGENE Census provides programmatic, versioned access to standardized single-cell genomics data from CZ CELLxGENE Discover. It contains 61+ million cells (human and mouse) with standardized metadata (cell types, tissues, diseases, donors), raw gene expression matrices, pre-calculated embeddings, and integration with PyTorch, scanpy, and other analysis tools.
When to Use This Skill
Use this skill when:
- Querying single-cell expression data by cell type, tissue, or disease
- Exploring available single-cell datasets and metadata
- Training machine learning models on single-cell data
- Performing large-scale cross-dataset analyses
- Integrating Census data with scanpy or other analysis frameworks
- Computing statistics across millions of cells
- Accessing pre-calculated embeddings or model predictions
For analyzing your own dataset (not the reference atlas), use scanpy or scvi-tools instead.
Installation
uv pip install cellxgene-census
uv pip install tiledbsoma-ml
Core Workflow
- Open the Census with a context manager; pin
census_version for reproducibility.
- Explore metadata first (
get_obs / datasets summary) to understand what's available — always filter is_primary_data == True to avoid duplicate cells.
- Estimate query size before loading expression. < 100k cells →
get_anndata() (in-memory); larger → axis_query() out-of-core iteration.
- Query expression with
obs_value_filter (cells) and var_value_filter (genes); select only the obs_column_names you need.
- Downstream: hand the returned AnnData to scanpy, or stream batches into a PyTorch dataloader for ML.
Minimal skeleton:
import cellxgene_census
with cellxgene_census.open_soma(census_version="2023-07-25") as census:
adata = cellxgene_census.get_anndata(
census=census,
organism="Homo sapiens",
obs_value_filter="cell_type == 'B cell' and tissue_general == 'lung' and is_primary_data == True",
)
Routing Guidance
- Small/medium query (fits in RAM) →
get_anndata(). See references/querying_expression.md.
- Query exceeds RAM →
axis_query() with chunked iteration and incremental stats. See references/querying_expression.md.
- Training ML models →
tiledbsoma_ml PyTorch dataloader / ExperimentDataset. See references/ml_and_scanpy.md.
- Standard scanpy analysis / multi-tissue integration → see
references/ml_and_scanpy.md.
- Need full schema, all metadata fields, or filter-syntax details →
references/census_schema.md.
Reference Index
references/querying_expression.md — Opening the Census, exploring metadata, small/medium get_anndata() queries, and large out-of-core axis_query() processing with incremental statistics.
references/ml_and_scanpy.md — tiledbsoma_ml PyTorch dataloader / ExperimentDataset train-test splits, scanpy integration, multi-dataset/tissue integration (anndata.concat), and four worked use cases.
references/best_practices_and_troubleshooting.md — Primary-data filtering, version pinning, query-size estimation, tissue_general vs tissue, presence matrices, the full obs/var metadata field list, and a troubleshooting guide.
references/census_schema.md — Census data structure, all metadata fields, value-filter syntax/operators, SOMA object types, and data inclusion criteria.
references/common_patterns.md — Extras beyond the core recipes: incremental (Welford) variance out-of-core, ontology-term filtering, batch-processing sweeps, and a common-pitfalls list.