cellxgene_standard / README.md
fvderop's picture
Update README.md
59c4b2f verified
# Cellxgene Human Aging Meta Summary
A curated summary of human single-cell RNA-seq datasets from [cellxgene](https://cellxgene.cziscience.com/), focused on aging and development, with standardized metadata across experiments and assays.
![Histogram](./assets/histogram.png)
## Intended Use
This summary is intended as a scaffold for integrating other curated or custom single-cell datasets (e.g., aging or disease-focused studies) with Cellxgene metadata. Enables exploration, QC, and pre-integration filtering at the experiment level. The `seafront` package contains code to pull the gene matrices associated with this metadata.
## Dataset Description
This dataset provides a summarized table where:
- Each row corresponds to a unique `dataset_id` + simplified assay combination (`experiment`).
- Numeric columns (e.g. `n_measured_vars`, `raw_sum`) are aggregated as **medians** per dataset.
- Categorical metadata (e.g. `cell_type`, `assay`, `tissue`, `disease`) are reported as **comma-separated unique values**.
- Ontology term IDs are also preserved and listed uniquely.
- Ambiguous developmental stages are filtered out; valid stages are converted to integer ages (`age_int`).
- Low-quality experiments (e.g. with median counts < 3000) are excluded.
## Data Sources
- Data pulled from the `cellxgene_census` API (version 2025-01) using `cellxgene_census.open_soma()`.
- All human single-cell datasets with interpretable metadata and sufficient quality metrics were included.
## Processing Steps
- **Assay Simplification**: Assay names are normalized into categories like `10x_3_prime`, `10x_5_prime`, `BD_WTA`, `ScaleBio`, etc.
- **Experiment Definition**: An `experiment` is defined as a unique combination of `dataset_id` and `assay_simple`.
- **Age Filtering**: Text-based developmental stages are mapped to integer age values using a rule-based filter.
- **Quality Filtering**: Datasets with median raw counts < 3000 are excluded.
- **Metadata Summary**: Output table generated using a per-experiment groupby summarization of all relevant columns.
## Files
- `cellxgene_human_dataset_overview.parquet`: Cleaned and summarized metadata per experiment.
- All variable names used in the datasets are saved for future integration compatibility.
## Example Columns
- `experiment`, `dataset_id`, `donor_id`
- `unique_cell_types_present`
- `unique_tissues_present`
- `unique_assays_present`, `assay_simple`
- `median_raw_sum`, `median_n_measured_vars`
- `age_int` (if available)
## Citation
Please cite the original Cellxgene datasets as per their respective DOIs. This metadata summary is derived and standardized but does not modify raw gene expression data.