---
# 1. Metadata Block
license: mit
task_categories:
- tabular-classification
tags:
- biology
- genomics
pretty_name: "gReLU Tutorial 2 dataset"
size_categories:
- 1M<n<10M
---

# binary-atac-tutorial-data

## Dataset Summary
This dataset contains binary accessibility values for 1154611 CREs in 222 cell types. The original source of this data is CATlas (https://decoder-genetics.wustl.edu/catlasv1/humanenhancer/data/cCRE_by_cell_type/). For more details, see https://decoder-genetics.wustl.edu/catlasv1/catlas_humanenhancer/#!/. This version of the dataset is used in gReLU tutorial 2 (https://github.com/Genentech/gReLU/blob/main/docs/tutorials/2_finetune.ipynb).

## Dataset Structure

### Data Fields
In `.obs`:

| Column | Type | Description |
| :--- | :--- | :--- |
| `cell type` | string | Name of the cell type |

In `.var`:

| Column | Type | Description |
| :--- | :--- | :--- |
| `chrom` | string | Chromosome (e.g., `chr6`) |
| `start` | int | Genomic start position (hg38) |
| `end` | int | Genomic end position (hg38) |
| `Class` | category | CRE class |
       
`.X` is a binary accessibility matrix of shape (222 × 1154611) in Compressed Sparse Row format.

## Usage

```python
import anndata
from huggingface_hub import hf_hub_download

file_path = hf_hub_download(
    repo_id="Genentech/binary-atac-tutorial-data",
    repo_type="dataset",
    filename="data.h5ad"
)
ad = anndata.read_h5ad(file_path)
```