AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

cellxgene-census

Queries the CZ CELLxGENE Census programmatically for versioned public single-cell and spatial transcriptomics data. Use when you need population-scale cell metadata, gene expression slices, Census summary counts, source H5AD URIs/downloads, embeddings, spatial Census data, or reference atlas comparisons across organisms, tissues, diseases, assays, and cell types. For analyzing your own local single-cell data use scanpy, anndata, or scvi-tools.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--cellxgene-census at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 5

references/common_patterns.md

Common Query Patterns and Best Practices

Reviewed against Census 1.18.0, TileDB-SOMA 2.3.0, and the 2025-11-08 LTS. Snippets share an open census context unless shown otherwise. Large cohort, model-training, and downstream analysis blocks are illustrative: first inspect metadata and size, restrict datasets/genes, and supply the named callbacks/model. API and sparse-statistics contracts were exercised on a small local SOMA fixture; a 10-cell/3-gene slice and metadata were also read from the public LTS.

Query Pattern Categories

1. Exploratory Queries (Metadata Only)

Use when exploring available data without loading expression matrices.

Pattern: Get unique cell types in a tissue

import cellxgene_census

with cellxgene_census.open_soma(census_version="2025-11-08") as census:
    cell_metadata = cellxgene_census.get_obs(
        census,
        "homo_sapiens",
        value_filter="tissue_general == 'brain' and is_primary_data == True",
        column_names=["cell_type"]
    )
    unique_cell_types = cell_metadata["cell_type"].unique()
    print(f"Found {len(unique_cell_types)} unique cell types")

Pattern: Count cells by condition

cell_metadata = cellxgene_census.get_obs(
    census,
    "homo_sapiens",
    value_filter="disease != 'normal' and is_primary_data == True",
    column_names=["disease", "tissue_general"]
)
counts = cell_metadata.groupby(["disease", "tissue_general"]).size()

Pattern: Explore dataset information

# Access datasets table
datasets = census["census_info"]["datasets"].read().concat().to_pandas()

# Title search is discovery only; disease labels are on obs, not datasets.
candidates = datasets[datasets["dataset_title"].str.contains("COVID", case=False, na=False)]

2. Small-to-Medium Queries (AnnData)

Use get_anndata() when results fit in memory. There is no safe universal cell-count threshold; include gene count, sparsity, embeddings, and downstream copies in the estimate.

Pattern: Tissue-specific cell type query

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",
    obs_column_names=["assay", "disease", "sex", "donor_id"],
)

Pattern: Gene-specific query with multiple genes

marker_genes = ["CD4", "CD8A", "CD19", "FOXP3"]

# First get gene IDs
gene_metadata = cellxgene_census.get_var(
    census, "homo_sapiens",
    value_filter=f"feature_name in {marker_genes}",
    column_names=["feature_id", "feature_name"]
)
gene_ids = gene_metadata["feature_id"].tolist()
if not gene_ids:
    raise ValueError("No requested marker genes found")
# Inspect duplicate symbols; retain feature_id rather than collapsing matches.

# Query with gene filter
adata = cellxgene_census.get_anndata(
    census=census,
    organism="Homo sapiens",
    var_value_filter=f"feature_id in {gene_ids}",
    obs_value_filter="cell_type == 'T cell' and is_primary_data == True",
)

Pattern: Multi-tissue query

adata = cellxgene_census.get_anndata(
    census=census,
    organism="Homo sapiens",
    obs_value_filter="tissue_general in ['lung', 'liver', 'kidney'] and is_primary_data == True",
    obs_column_names=["cell_type", "tissue_general", "dataset_id"],
)

Pattern: Disease-specific query

adata = cellxgene_census.get_anndata(
    census=census,
    organism="Homo sapiens",
    obs_value_filter="disease == 'COVID-19' and tissue_general == 'lung' and is_primary_data == True",
)

Pattern: Include compound disease labels

Schema 2.4.0 stores multiple disease labels/IDs in one string separated by ||. Use summary rows to discover the actual strings and token membership to choose whole values; substring matching can confuse distinct disease names. This includes co-diagnoses rather than only the exact single-label cohort.

summary = census["census_info"]["summary_cell_counts"].read().concat().to_pandas()
disease_values = summary.loc[
    summary["organism"].eq("homo_sapiens") & summary["category"].eq("disease"),
    "label",
].unique()
matching = [value for value in disease_values if "COVID-19" in value.split(" || ")]
if not matching:
    raise ValueError("Requested disease not present in this release")
disease_filter = f"disease in {matching!r} and is_primary_data == True"
# Add dataset/tissue restrictions after inspecting cohort size.

For precise ontology-based cohorts, inspect disease_ontology_term_id as well; matching one ontology term does not automatically include all descendants.

3. Large Queries (Out-of-Core Processing)

Use axis_query() for queries that exceed available RAM.

Pattern: Iterative processing

import tiledbsoma as soma

# Create query
with census["census_data"]["homo_sapiens"].axis_query(
    measurement_name="RNA",
    obs_query=soma.AxisQuery(
        value_filter="tissue_general == 'brain' and is_primary_data == True"
    ),
    var_query=soma.AxisQuery(
        value_filter="feature_name in ['FOXP2', 'TBR1', 'SATB2']"
    ),
) as query:
    # Iterate through X matrix in chunks
    iterator = query.X("raw").tables()
    for batch in iterator:
        # Process batch (a pyarrow.Table); soma_dim_0 = cells, soma_dim_1 = genes
        # batch has columns: soma_data, soma_dim_0, soma_dim_1
        process_batch(batch)

Pattern: Per-gene mean and variance including measured zeros

Restrict to datasets that measured every requested feature before computing the statistics. This example selects CD4 by its unique Ensembl ID. It includes sparse zeros, returns one row per gene, and rejects empty or one-cell selections rather than returning an undefined sample variance.

from cellxgene_census.experimental.pp import mean_variance

# Assumes census and soma are imported/open as above.
genes = cellxgene_census.get_var(
    census, "homo_sapiens", value_filter="feature_id == 'ENSG00000010610'",
    column_names=["soma_joinid", "feature_id"],
)
if len(genes) != 1:
    raise ValueError("Expected one CD4 feature")
datasets = census["census_info"]["datasets"].read().concat().to_pandas()
presence = cellxgene_census.get_presence_matrix(census, "homo_sapiens")
covered = presence[datasets["soma_joinid"].to_numpy(), :][:, genes["soma_joinid"].to_numpy()]
eligible_ids = datasets.loc[covered.toarray().all(axis=1), "dataset_id"].tolist()
if not eligible_ids:
    raise ValueError("No datasets measured CD4")

with census["census_data"]["homo_sapiens"].axis_query(
    measurement_name="RNA",
    obs_query=soma.AxisQuery(value_filter=(
        f"dataset_id in {eligible_ids!r} and tissue_general == 'brain' "
        "and is_primary_data == True"
    )),
    var_query=soma.AxisQuery(coords=(genes["soma_joinid"].to_numpy(),)),
) as query:
    if query.n_obs < 2:
        raise ValueError("At least two selected cells are required")
    stats = mean_variance(
        query, layer="raw", axis=0, calculate_mean=True,
        calculate_variance=True, ddof=1, nnz_only=False,
    )

These describe raw count distributions, not normalized differential expression. For genes with different measurement coverage, either use a common measured subset or compute each gene with its own valid denominator. nnz_only=True is a different estimand: the distribution of explicitly stored entries.

4. PyTorch Integration (Machine Learning)

Use TileDB-SOMA-ML for training models. The former cellxgene_census.experimental.ml loaders are absent from 1.18.0.

Pattern: Create training dataloader

import tiledbsoma as soma
from tiledbsoma_ml import ExperimentDataset, experiment_dataloader

with cellxgene_census.open_soma(census_version="2025-11-08") as census:
    experiment = census["census_data"]["homo_sapiens"]
    with experiment.axis_query(
        measurement_name="RNA",
        obs_query=soma.AxisQuery(
            value_filter="tissue_general == 'liver' and is_primary_data == True"
        ),
    ) as query:
        dataset = ExperimentDataset(
            query=query,
            layer_name="raw",
            obs_column_names=["cell_type"],
            batch_size=128,
            shuffle=True,
        )
        dataloader = experiment_dataloader(dataset)

        for epoch in range(num_epochs):
            dataset.set_epoch(epoch)
            for X, obs in dataloader:
                labels = obs["cell_type"]
                # Train model...

Pattern: Train/test split

# Split data
train_dataset, test_dataset = dataset.random_split(0.8, 0.2, seed=42)

# Create loaders
train_loader = experiment_dataloader(train_dataset, num_workers=0)
test_loader = experiment_dataloader(test_dataset, num_workers=0)

Set batch_size and shuffle on ExperimentDataset, not on the PyTorch DataLoader. On macOS, enable multiple workers only from an importable script under if __name__ == "__main__":. Keep iteration within the open Census/query context.

A random cell split measures within-cohort interpolation; hold out donors or studies to test generalization. Keep one vocabulary fitted on the training data and define how unseen labels are handled. X is NumPy (or SciPy sparse), and obs is pandas, so convert inputs to tensors and string labels to stable integer IDs before computing a PyTorch loss.

5. Spatial Census Data

Use the cellxgene-census[spatial] extra and query the census_spatial_sequencing collection for Visium or Slide-seq V2 data.

import tiledbsoma as soma

with cellxgene_census.open_soma(census_version="2025-11-08") as census:
    spatial_experiment = census["census_spatial_sequencing"]["homo_sapiens"]
    with spatial_experiment.axis_query(
        measurement_name="RNA",
        obs_query=soma.AxisQuery(
            value_filter="dataset_id == '4cceac62-9513-42a4-90e5-2878dbb0192c'"
        ),
    ) as query:
        sdata = query.to_spatialdata(X_name="raw")

6. Integration Workflows

Pattern: Scanpy integration

import scanpy as sc

# Load data
adata = cellxgene_census.get_anndata(
    census=census,
    organism="Homo sapiens",
    obs_value_filter="cell_type == 'neuron' and is_primary_data == True",
)

# Standard scanpy workflow
sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata)
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.umap(adata)
sc.pl.umap(adata, color=["cell_type", "tissue_general"])

Pattern: Multi-dataset integration

# Query multiple datasets separately
datasets_to_integrate = ["dataset_id_1", "dataset_id_2", "dataset_id_3"]

adatas = []
for dataset_id in datasets_to_integrate:
    adata = cellxgene_census.get_anndata(
        census=census,
        organism="Homo sapiens",
        obs_value_filter=f"dataset_id == '{dataset_id}' and is_primary_data == True",
    )
    adatas.append(adata)

# Scanpy's wrapper expects ONE AnnData with contiguous batches and X_pca.
# Install scanorama separately for this illustrative integration workflow.
import anndata as ad
import scanpy as sc
import scanpy.external as sce
for item in adatas:
    item.var_names = item.var["feature_id"].astype(str)
combined = ad.concat(adatas, label="batch", keys=datasets_to_integrate, join="inner")
combined.layers["counts"] = combined.X.copy()
sc.pp.normalize_total(combined, target_sum=1e4)
sc.pp.log1p(combined)
sc.pp.highly_variable_genes(combined, n_top_genes=2000, batch_key="batch")
sc.pp.pca(combined)
sce.pp.scanorama_integrate(combined, key="batch")
sc.pp.neighbors(combined, use_rep="X_scanorama")

Best Practices

1. Always Filter for Primary Data

Unless specifically analyzing duplicates, always include is_primary_data == True:

obs_value_filter="cell_type == 'B cell' and is_primary_data == True"

2. Specify Census Version

For reproducible analysis, always specify the Census version:

census = cellxgene_census.open_soma(census_version="2025-11-08")

3. Use Context Manager

Always use the context manager to ensure proper cleanup:

with cellxgene_census.open_soma(census_version="2025-11-08") as census:
    print(list(census["census_data"]))

4. Select Only Needed Columns

Minimize data transfer by selecting only required metadata columns:

obs_column_names=["cell_type", "tissue_general", "disease"]  # Not all columns

5. Check Dataset Presence for Gene Queries

When analyzing specific genes, check which datasets measured them:

genes = cellxgene_census.get_var(
    census, "homo_sapiens",
    value_filter="feature_name in ['CD4', 'CD8A']",
    column_names=["soma_joinid", "feature_id", "feature_name"],
)
presence = cellxgene_census.get_presence_matrix(census, "homo_sapiens")
# Columns use Census join IDs, not positions in the filtered gene table.
gene_presence = presence[:, genes["soma_joinid"].to_numpy()]

6. Use tissue_general for Broader Queries

tissue_general provides coarser groupings than tissue, useful for cross-tissue analyses:

# Better for broad queries
obs_value_filter="tissue_general == 'immune system'"

# Use specific tissue when needed
obs_value_filter="tissue == 'venous blood'"

7. Combine Metadata Exploration with Expression Queries

First explore metadata to understand available data, then query expression:

# Step 1: Explore
metadata = cellxgene_census.get_obs(
    census, "homo_sapiens",
    value_filter="disease == 'COVID-19'",
    column_names=["cell_type", "tissue_general"]
)
print(metadata.value_counts())

# Step 2: Query based on findings
adata = cellxgene_census.get_anndata(
    census=census,
    organism="Homo sapiens",
    obs_value_filter="disease == 'COVID-19' and cell_type == 'T cell' and is_primary_data == True",
)

8. Memory Management for Large Queries

For large queries, check estimated size before loading:

# Get cell count first
metadata = cellxgene_census.get_obs(
    census, "homo_sapiens",
    value_filter="tissue_general == 'brain' and is_primary_data == True",
    column_names=["soma_joinid"]
)
n_cells = len(metadata)
print(f"Query will return {n_cells} cells")

# If too large, use out-of-core processing or further filtering

9. Leverage Ontology Terms for Consistency

When possible, use ontology term IDs instead of free text:

# More reliable than cell_type == 'B cell' across datasets
obs_value_filter="cell_type_ontology_term_id == 'CL:0000236'"

10. Batch Processing Pattern

For systematic analyses across multiple conditions:

tissues = ["lung", "liver", "kidney", "heart"]
results = {}

for tissue in tissues:
    adata = cellxgene_census.get_anndata(
        census=census,
        organism="Homo sapiens",
        obs_value_filter=f"tissue_general == '{tissue}' and is_primary_data == True",
    )
    # Perform analysis
    results[tissue] = analyze(adata)

Common Pitfalls to Avoid

  1. Not filtering for is_primary_data: Leads to counting duplicate cells
  2. Loading too much data: Use metadata queries to estimate size first
  3. Not using context manager: Can cause resource leaks
  4. Inconsistent versioning: Results not reproducible without specifying version
  5. Overly broad queries: Start with focused queries, expand as needed
  6. Ignoring dataset presence: Some genes not measured in all datasets
  7. Wrong count normalization: Be aware of UMI vs read count differences