README.md

---
license: mit
task_categories:
- tabular-classification
tags:
- biology
- genomics
pretty_name: "CATlas Human Enhancer Binary Matrix"
size_categories:
- 1M<n<10M
---

# human-catlas-atac-data

## Dataset Summary
This dataset provides a binary accessibility matrix of candidate Cis-Regulatory Elements (cCREs) across human cell types. It is derived from the CATlas project (https://decoder-genetics.wustl.edu/catlasv1/catlas_humanenhancer/#!/). The matrix identifies which genomic regions are accessible (1) or inaccessible (0) across 204 distinct cell types. The data is derived from https://decoder-genetics.wustl.edu/catlasv1/humanenhancer/data/cCRE_by_cell_type/. The hg38 genome build was used.

## Repository Content
1. `data.h5ad`: The main dataset stored in AnnData format.
2. `1_data.ipynb`: Jupyter notebook containing the preprocessing steps used to generate the `.h5ad` file.

## Dataset Structure

### AnnData Object Dimensions
- **n_obs (Cell Types):** 204
- **n_vars (cCREs):** 1,121,319

### Data Fields
- **`.X`**: A sparse binary matrix where rows are cell types and columns are genomic regions (cCREs).
- **`.obs` (Cell Type Metadata):**
  - `cell type`: The descriptive name of the human cell type/cluster.
- **`.var` (Genomic Feature Metadata):**
  - `chrom`, `start`, `end`: Genomic coordinates (hg38).
  - `cre_class`: Classification of the regulatory element.
  - `in_fetal` / `in_adult`: indicators of activity in developmental stages.
  - `cre_module`: Associated regulatory module.
  - `enformer_split`: Overlap with the data splits used for training the Enformer model.
  - `split`: Splits used for downstream modeling (training/validation/test).

## Usage

```python
import anndata as ad
from huggingface_hub import hf_hub_download

file_path = hf_hub_download(
    repo_id="Genentech/human-atac-catlas-data", 
    filename="data.h5ad"
)

ad = anndata.read_h5ad(file_path)
```