cellxgene-context

Integrating CellxGene Census Single-Cell Data with ENCODE Bulk Experiments

Bridge bulk ENCODE functional genomics data with cell-type-specific expression from the CellxGene Census, the largest unified single-cell RNA-seq atlas, to resolve cell-type contributions to regulatory element activity.

Scientific Rationale

The question: "Which specific cell types within my tissue drive the regulatory signals I see in bulk ENCODE data?"

ENCODE provides deeply sequenced bulk functional genomics (ChIP-seq, ATAC-seq, Hi-C) across hundreds of biosamples. But bulk data from a tissue like "pancreas" is a mixture of acinar cells (~80%), duct cells (~10%), endocrine cells (~5%), and others. An H3K27ac peak in bulk pancreas could be driven by any of these cell types. CellxGene Census provides cell-type-resolved expression data from 50M+ single-cell observations across thousands of datasets, enabling deconvolution of bulk ENCODE signals.

The Bulk-to-Single-Cell Bridge

| Bulk ENCODE Signal | Single-Cell Question | CellxGene Answer | |-------------------|---------------------|-----------------| | H3K27ac peak near INS gene in pancreas | Which cell type expresses INS? | Beta cells (>500 TPM), not acinar (<1 TPM) | | ATAC-seq peak in liver near ALB | Is this hepatocyte-specific? | Yes — ALB expressed only in hepatocytes | | Enhancer active in brain cortex | Neurons or glia? | CellxGene resolves excitatory neurons vs. astrocytes vs. oligodendrocytes | | Broad H3K27ac domain in blood | Which immune cell type? | Can distinguish T cells, B cells, monocytes, NK cells |

What CellxGene Census Provides

• 50M+ single-cell observations from thousands of published datasets
• Standardized cell ontology (Cell Ontology terms) across all datasets
• Unified gene expression in a consistent format
• Metadata: tissue, disease status, sex, ethnicity, developmental stage
• API access via Python (cellxgene-census) or R (cellxgene.census)
• No authentication required for public data

Key Literature

• Megill et al. 2021 "cellxgene: a performant, scalable exploration platform for high dimensional sparse matrices" (bioRxiv preprint). Describes the CellxGene platform architecture and exploration capabilities. DOI: 10.1101/2021.04.05.438318
• CZ CELLxGENE Discover (Chan Zuckerberg Initiative, 2023). CellxGene Census provides programmatic access to the entire CellxGene data corpus as a single unified dataset. https://cellxgene.cziscience.com/
• Regev et al. 2017 "The Human Cell Atlas" (eLife, ~1,500 citations). The vision paper for comprehensive single-cell reference maps of all human cells. CellxGene Census is the largest realization of this vision. DOI: 10.7554/eLife.27041
• Tabula Sapiens Consortium 2022 "The Tabula Sapiens: A multiple-organ, single-cell transcriptomic atlas of humans" (Science, ~800 citations). Multi-organ human cell atlas contributing to CellxGene Census. DOI: 10.1126/science.abl4896
• ENCODE Project Consortium 2020 (Nature, ~1,656 citations). The bulk regulatory element catalog that CellxGene single-cell data contextualizes. DOI: 10.1038/s41586-020-2493-4

When to Use This Skill

| Scenario | How CellxGene Helps | |---------|-------------------| | Bulk ENCODE peak near a gene — which cell type? | Query gene expression by cell type in matching tissue | | ENCODE enhancer active in tissue X — cell-type-specific? | Check if enhancer target gene is restricted to one cell type | | Choosing ENCODE cell line as proxy | Verify which primary cell type the cell line best represents | | Interpreting differential peaks between tissues | Determine if difference is due to cell-type composition | | Validating ENCODE scATAC-seq findings | Cross-reference with CellxGene scRNA-seq for same cell types | | Designing follow-up experiments | Identify which cell types to isolate for validation |

Python API Reference

Installation

pip install cellxgene-census

Requires Python 3.8+. The package uses TileDB-SOMA for efficient data access.

Core API Pattern

import cellxgene_census

# Open the Census (reads metadata, does not download all data)
with cellxgene_census.open_soma() as census:
    # Access human data
    human = census["census_data"]["homo_sapiens"]

    # Query specific genes in specific tissues/cell types
    # This is where filtering happens — be specific to control memory

Step 1: Identify the ENCODE Target Gene

Start from an ENCODE finding — a regulatory element near a gene of interest:

# Find enhancers in pancreas
encode_search_experiments(
    assay_title="Histone ChIP-seq",
    target="H3K27ac",
    organ="pancreas",
    biosample_type="tissue"
)

# Get peaks
encode_list_files(
    experiment_accession="ENCSR...",
    file_format="bed",
    output_type="IDR thresholded peaks",
    assembly="GRCh38"
)

From peaks, identify the nearest gene(s). You need the gene symbol or Ensembl ID.

Step 2: Query CellxGene Census for Cell-Type Expression

Basic Gene Expression Query

import cellxgene_census
import pandas as pd

gene_symbol = "INS"  # Insulin — example for pancreas

with cellxgene_census.open_soma() as census:
    human = census["census_data"]["homo_sapiens"]

    # Get expression for INS in pancreas tissue
    # Use obs_value_filter to restrict to pancreas
    # Use var_value_filter to restrict to the gene
    adata = cellxgene_census.get_anndata(
        census,
        organism="Homo sapiens",
        var_value_filter=f"feature_name == '{gene_symbol}'",
        obs_value_filter="tissue_general == 'pancreas'",
        obs_column_names=["cell_type", "tissue", "disease", "dataset_id"]
    )

    # Summarize expression by cell type
    expr_by_celltype = adata.to_df().join(adata.obs["cell_type"])
    summary = expr_by_celltype.groupby("cell_type").agg(
        mean_expr=(gene_symbol, "mean"),
        pct_expressed=(gene_symbol, lambda x: (x > 0).mean() * 100),
        n_cells=(gene_symbol, "count")
    ).sort_values("mean_expr", ascending=False)

    print(summary.head(10))

Multi-Gene Query

genes_of_interest = ["INS", "GCG", "SST", "PPY"]  # Islet hormones

with cellxgene_census.open_soma() as census:
    gene_filter = " or ".join([f"feature_name == '{g}'" for g in genes_of_interest])

    adata = cellxgene_census.get_anndata(
        census,
        organism="Homo sapiens",
        var_value_filter=gene_filter,
        obs_value_filter="tissue_general == 'pancreas'",
        obs_column_names=["cell_type", "tissue", "disease"]
    )

Query by Cell Ontology Term

# More precise than tissue — query specific cell types
with cellxgene_census.open_soma() as census:
    adata = cellxgene_census.get_anndata(
        census,
        organism="Homo sapiens",
        var_value_filter="feature_name == 'INS'",
        obs_value_filter="cell_type == 'type B pancreatic cell'",  # Cell Ontology term for beta cells
        obs_column_names=["cell_type", "tissue", "disease", "sex"]
    )

Step 3: Map CellxGene Cell Types to ENCODE Biosamples

CellxGene uses Cell Ontology (CL) terms. ENCODE uses its own biosample ontology. Key mappings:

| CellxGene Cell Type (CL term) | ENCODE Biosample | Notes | |-------------------------------|-----------------|-------| | type B pancreatic cell | pancreatic beta cell | Beta cells | | hepatocyte | hepatocyte | Direct match | | CD4-positive, alpha-beta T cell | CD4+ T cell | ENCODE may have more specific subtypes | | monocyte | monocyte | Direct match | | excitatory neuron | neuron | ENCODE may use broader category | | oligodendrocyte | oligodendrocyte | Direct match | | fibroblast | fibroblast | Direct match | | endothelial cell | endothelial cell of umbilical vein (HUVEC) |