🧬 Aging Wiki

finding-singlecell-aging

Properties1
aliasessingle-cell aging, scRNA-seq, Tabula Muris Senis, CellxGene, TMS, aging atlas single-cell

7 min read

SOP: finding single-cell aging data

Use this SOP to find and interpret single-cell transcriptomic (scRNA-seq) data on cell-type-specific changes during aging. The key resources are Tabula Muris Senis (mouse, gold-standard multi-tissue aging atlas), CellxGene Census (multi-study human/mouse aggregator), and the Aging Atlas CNCB single-cell tier. Results from this SOP populate single-cell-aging-signature: on cell-type pages and inform “Cell-type aging” subsections on protein/process pages.

When to use single-cell data

  • Writing or extending a type: cell-type page — the cell-type’s aging signature belongs in frontmatter (single-cell-aging-signature:) and in the body.
  • Checking whether a bulk tissue expression-with-age claim is cell-composition confound vs genuine per-cell change (see also sops/finding-tissue-expression.md).
  • Describing which cell types in a tissue show the strongest aging-associated transcriptional changes.
  • Validating that a protein implicated in aging (from mouse knockout or bulk transcriptomics) shows age-correlated expression specifically in relevant cell types.

Resource 1: Tabula Muris Senis (TMS)

What it is: The mouse single-cell aging atlas. scRNA-seq and snRNA-seq of ~350K cells from 23 tissues collected from C57BL/6J mice at 1, 3, 18, 21, 24, and 30 months. Covers 120+ cell types. The authoritative reference for “what happens in cell type X with age in mouse.”

Primary paper: Tabula Muris Consortium (2020), Nature — DOI: 10.1038/s41586-020-2496-1.

Data access:

Access routeURL / detail
Interactive browserhttps://tabula-muris-senis.ds.czbiohub.org/
Figshare DOI cluster10.6084/m9.figshare.8273102 (plus companion figshare IDs per tissue)
AWS Open Datas3://czb-tabula-muris-senis/ (free, requester-pays for egress)
CellxGene ExplorerHosted collections via CellxGene Census (see Resource 2)

Figshare datasets — key IDs:

  • 10.6084/m9.figshare.8273102 — full atlas (10Ă— and Smart-Seq2, combined)
  • Individual tissue h5ad files downloadable from figshare; file size 50–500 MB each

Programmatic access (recommended for scripting):

import cellxgene_census
import tiledb
 
# Open the Census (includes TMS data under mus_musculus)
census = cellxgene_census.open_soma(census_version="stable")
 
# Query TMS heart cells from aged vs young mice
obs_df = census["census_data"]["mus_musculus"].obs.read(
    value_filter="tissue == 'heart' and dataset_title =~ '.*Tabula Muris Senis.*'",
    column_names=["cell_type", "development_stage", "sex", "dataset_title"]
).concat.to_pandas
 
census.close

Key metadata fields in TMS obs:

  • tissue — tissue of origin (anatomical term)
  • age — donor age in months (1, 3, 18, 21, 24, 30)
  • cell_ontology_class — CL: ontology cell type
  • free_annotation — fine-grained curator annotation
  • method — 10x or smartseq2

What to extract for a cell-type page:

  1. Differentially expressed genes between young (3 months) and old (24 months) in the target cell type — top 10 up/down.
  2. Whether the cell type shows transcriptional “aging score” changes (Tabula Muris Senis companion paper reports per-tissue and per-cell-type aging scores using gene-module scoring).
  3. Cell-composition changes: does the proportion of this cell type change with age in its tissue?

Caveat: TMS is mouse only. All TMS findings should carry the standard extrapolation table (see CLAUDE.md) before inferring human equivalents.

Resource 2: CellxGene Census (Chan-Zuckerberg Initiative)

What it is: A unified, harmonized aggregation of curated scRNA-seq datasets from CZ CELLxGENE Discover. As of 2025, contains >80 million human cells and >20 million mouse cells from hundreds of datasets, all in standardized AnnData/SOMA format. Supports cross-dataset queries.

Primary API: https://api.cellxgene.cziscience.com/curation/v1/ (dataset metadata) Python SDK (recommended): cellxgene_census package (PyPI)

Dataset discovery

# List all datasets, paginated (returns metadata including title, organism, tissues, schema)
curl "https://api.cellxgene.cziscience.com/curation/v1/datasets?limit=100"

Confirmed response fields (validated 2026-05-05):

  • dataset_id, dataset_version_id
  • title — human-readable dataset name
  • cell_count, primary_cell_count, mean_genes_per_cell
  • assay — e.g., [{"label": "10x 3' v3"}]
  • tissue — array of ontology terms
  • organism — Homo sapiens / Mus musculus
  • development_stage — age/developmental stage of donors
  • disease — disease context (normal vs disease)
  • cell_type — array of cell types
  • explorer_url — direct link to CellxGene explorer for this dataset

Programmatic cell-type query

import cellxgene_census
 
census = cellxgene_census.open_soma(census_version="stable")
 
# Human aged tissue (kidney, muscle, brain) cells
obs_df = census["census_data"]["homo_sapiens"].obs.read(
    value_filter=(
        "tissue_general in ['kidney', 'skeletal muscle tissue', 'brain'] "
        "and disease == 'normal'"
    ),
    column_names=["cell_type", "development_stage", "donor_id", "tissue", "dataset_id"]
).concat.to_pandas
 
# Summarize by age bracket
obs_df["age_group"] = obs_df["development_stage"].str.extract(r"(\d+)")
 
census.close

Key metadata fields in Census obs:

  • tissue_general — coarse tissue (e.g., kidney)
  • tissue — fine tissue (e.g., kidney cortex)
  • cell_type — CL ontology cell type string
  • development_stage — HsapDv ontology term (e.g., “70-year-old human stage”)
  • disease — EFO ontology (filter to normal for aging comparisons)
  • sex, self_reported_ethnicity
  • donor_id — for grouping per-donor

Age filtering in Census: Human aging datasets use HsapDv ontology development stage terms. For filtering to aged adults, use:

# ages 60–89 years
value_filter = "development_stage in ['HsapDv:0000246', 'HsapDv:0000247', 'HsapDv:0000248']"

Alternatively filter on a string pattern: development_stage =~ ".*[6-9][0-9]-year.*".

Resource 3: Aging Atlas CNCB — single-cell tier

What it is: The CNCB Aging Atlas (https://ngdc.cncb.ac.cn/aging/) curates published single-cell aging datasets with standardized metadata and interactive visualization. Includes human and mouse scRNA-seq studies focused on aging phenotypes.

Note: The Aging Atlas does not expose a live REST API (confirmed 2026-05-05). Data is accessed via the web portal and bulk downloads from the CNCB FTP.

When to use: For finding curated aging-focused scRNA datasets not in CellxGene Census. The Aging Atlas editorial curation is tighter (aging studies only); Census is broader but less aging-focused.

Citing Aging Atlas: Paper DOI 10.1093/nar/gkaa894 (Zhang et al. 2020, NAR). Include database version in citation.

The single-cell-aging-signature: frontmatter field

This R22-introduced field on type: cell-type pages encodes the cell type’s aging signature as a free-text summary:

single-cell-aging-signature: "TMS 24m vs 3m: upregulated p21, p16, Ccl2; downregulated Myh7, Tnni3; cell proportion stable; data from Tabula Muris Senis 2020"

Encoding convention:

  • Start with data source and age comparison (e.g., TMS 24m vs 3m:)
  • List top 3–5 upregulated genes (gene symbols, mouse or human)
  • List top 3–5 downregulated genes
  • Note whether cell proportions change with age
  • End with the data source citation
  • Use null + #gap/no-singlecell-data if not yet populated

Full table: For cell-type pages with extensive data, also add a body table with DEGs, effect sizes, and the species/tissue context. The frontmatter field is a quick-access summary only.

Interpreting single-cell aging data — caveats

Batch effects across studies: Cross-dataset comparisons in CellxGene Census require batch correction. Genes that appear up in one study’s “aged” condition may reflect technical batch rather than biology. Prefer within-study comparisons (e.g., TMS which controls technology and genetic background).

Cell annotation heterogeneity: Cell type labels differ across studies. “Macrophage” in one study may be “monocyte-derived macrophage” in another. Use CL ontology terms where available for cross-study comparability.

Pseudobulk for DEG: Single-cell DEG analysis is best done pseudobulk (aggregate per-cell-type per-donor, then use bulk DEG methods). Naive single-cell tests inflate N and produce spurious significance. When citing scRNA-seq DEGs, note whether the paper used pseudobulk or naive approaches.

Mouse vs human: All TMS data is mouse. Map findings to human where possible via Census human data or HPA single-cell. Flag any claim extrapolated from mouse scRNA without human support as #gap/needs-human-replication.

Tissue composition vs per-cell expression: Bulk RNA-seq “age-correlated expression” is often explained by cell-type proportion shifts (e.g., more senescent cells in aged tissue). Single-cell data disambiguates this. When citing bulk tissue aging claims, always check if single-cell data is available to rule out cell-composition confound.

Workflow for a cell-type page

  1. Search CellxGene Census for datasets containing this cell type across ages in both human and mouse.
  2. Query TMS for the mouse equivalent — extract top DEGs (young vs old) if the cell type is represented in TMS.
  3. Cross-check top DEGs against the protein pages already in the wiki (do any implicated genes have pages here?).
  4. Summarize cell-proportion changes (is this cell type more/less abundant in aged tissue?).
  5. Fill single-cell-aging-signature: frontmatter.
  6. Write a “Single-cell aging evidence” section in the body with citations.
  7. If no TMS representation and no Census data: set field to null + #gap/no-singlecell-data.

Linking single-cell data to protein pages

On type: protein pages, under a “Cell-type-specific expression” heading, reference relevant cell-type pages (via [[wikilinks]]) and note whether TMS or Census data shows age-correlated expression of this protein specifically in one cell type.

See also

Graph View

Backlinks