Genentech/borzoi-data

---
license: mit
task_categories:
- tabular-regression
tags:
- biology
- genomics
pretty_name: "Borzoi Intervals"
size_categories:
- 100K<n<1M
---

# borzoi-data

## Dataset Summary
This dataset contains the specific genomic intervals used for training, validating, and testing the Borzoi model, a deep learning architecture for predicting functional genomic tracks from DNA sequence. The intervals are provided for both human and mouse genomes. We modified the intervals provided in the original source by extending the input sequence to 524,288 bp to create the full interval that was supplied to the model.

- **Publication:** [Linder, J. et al. Nature Genetics (2025)](https://www.nature.com/articles/s41588-024-02053-6)
- **Original Source:** https://github.com/calico/borzoi/tree/main/data
- **Genome Builds:**
- Human: hg38
- Mouse: mm10

## Repository Content
The repository includes two tab-separated values (TSV) files and two Jupyter notebooks:
1. `human_intervals.tsv`: 55,497 genomic regions (excluding header).
2. `mouse_intervals.tsv`: 49,369 genomic regions (excluding header).
3. `data_human.ipynb`: Code to create `human_intervals.tsv`.
4. `data_mouse.ipynb`: Code to create `mouse_intervals.tsv`.

## Dataset Structure

### Data Fields
Both files follow a standard genomic interval format:

| Column | Type | Description |
| :--- | :--- | :--- |
| `chrom` | string | Chromosome identifier (e.g., `chr18`, `chr4`) |
| `start` | int | Start coordinate of the interval |
| `end` | int | End coordinate of the interval |
| `fold` | string | Fold assignment (`fold0`-`fold7`) |
| `split` | string | Data partition assignment (`train`, `test`, or `val`) |

### Statistics
| File | Number of Regions | Genome Build |
| :--- | :--- | :--- |
| `human_intervals.tsv` | 55,497 | hg38 |
| `mouse_intervals.tsv` | 49,369 | mm10 |

### Data Splits

**Human (hg38)**
| Split | Count |
|-------|-------|
| train | 41,699 |
| val | 6,910 |
| test | 6,888 |

**Mouse (mm10)**
| Split | Count |
|-------|-------|
| train | 36,950 |
| val | 6,318 |
| test | 6,101 |

## Usage

```python
from huggingface_hub import hf_hub_download
import pandas as pd

file_path = hf_hub_download(
repo_id="Genentech/borzoi-data",
filename="human_intervals.tsv",
repo_type="dataset"
)
df_human = pd.read_csv(file_path, sep='\t')

file_path = hf_hub_download(
repo_id="Genentech/borzoi-data",
filename="mouse_intervals.tsv",
repo_type="dataset"
)
df_mouse = pd.read_csv(file_path, sep='\t')
```