---
title: ChromBPNet
canonical_url: "https://www.modelscope.cn/datasets/User040218/ChromBPNet"
md_url: "https://www.modelscope.cn/datasets/User040218/ChromBPNet.md"
repository: User040218/ChromBPNet
last_updated: 2026-11-01
license: cc-by-4.0
storage_size: "43 GB"
downloads: 25296
stars: 0
---

# ChromBPNet

> ChromBPNet - User040218 在 ModelScope 开源的数据集。ChromBPNet: 2,114 bp sequences and base-resolution accessibility profiles from 9 ENCODE ATAC/DNase experiments (5 folds), plus the tutorial data.

User040218/ChromBPNet 是 ModelScope 魔搭社区上的数据集，存储大小 43 GB，采用 cc-by-4.0 许可。

- **Repository**: User040218/ChromBPNet
- **License**: cc-by-4.0
- **Storage size**: 43 GB
- **Downloads**: 25296
- **Stars**: 0
- **Last updated**: 2026-11-01

Source: https://www.modelscope.cn/datasets/User040218/ChromBPNet

---

# ChromBPNet

**2,114 bp sequences with base-resolution accessibility profiles from 9 ENCODE ATAC-seq/DNase-seq experiments (5 folds)** — regression.

ChromBPNet models chromatin accessibility (ATAC-seq / DNase-seq) at base resolution after correcting for enzyme bias. This release contains the 9 ENCODE experiments used in the paper, each with the authors' 5 chromosome folds (fold_0–fold_4) and train/validation/test partitions, plus the binary task of the official tutorial (`tutorial_binary`).

Each example is a 2,114 bp DNA sequence; the target is the per-base read count over the central 1,000 bp (1,000 values), and `log1p_total_counts` holds the total count.

| Experiment | Cell line | Assay |
|---|---|---|
| ENCSR003WJE | GM12878 | DNase-seq |
| ENCSR389HIH | GM12878 | ATAC-seq |
| ENCSR006CUK | HepG2 | DNase-seq |
| ENCSR380YGX | HepG2 | ATAC-seq |
| ENCSR296UHQ | K562 | DNase-seq |
| ENCSR467RSV | K562 | ATAC-seq |
| ENCSR137OLC | IMR-90 | DNase-seq |
| ENCSR978WIX | IMR-90 | ATAC-seq |
| ENCSR085MTT | H1 | DNase-seq |

Links: [ChromBPNet (GitHub)](https://github.com/kundajelab/chrombpnet).

## Subsets (46)

All 46 previewable subsets are row-level indices named `index_<experiment>_fold_<k>` and `index_tutorial_binary`; they record coordinates, annotations, fold membership and native file locations. The same region appears in different folds and should not be counted as independent samples. Training data are in the native files.

46 subsets, 26,215,128 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `sequence` | — | 2,114 bp DNA sequence (native files) |
| `labels` | — | Per-base counts over the central 1,000 bp (native files) |
| `log1p_total_counts` | float64 | log(1 + total counts) |
| `chrom` / `center` | string / int64 | Region chromosome and center |
| `fold_0 … fold_4` | string | Fold membership |

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `ENCSR003WJE_fold_0`, regression, `num_labels=1000`.

```python
import json
from modelscope import snapshot_download
from dnallm import DNADataset

repo = "User040218/ChromBPNet"
root = snapshot_download(repo_id=repo, repo_type="dataset", allow_patterns=["source/native-file-map.json"])
files = json.load(open(f"{root}/source/native-file-map.json", encoding="utf-8"))["ENCSR003WJE_fold_0"]
root = snapshot_download(repo_id=repo, repo_type="dataset",
                         allow_patterns=[f["path"] for split in files.values() for f in split])
data = DNADataset.load_local_data(
    {split: [f"{root}/{f['path']}" for f in items] for split, items in files.items()},
    seq_col="sequence", label_col="labels", max_length=2116)
```

The index subsets shown in the web preview are for browsing and locating rows only; the code above downloads and reads the native files listed in `source/native-file-map.json`.

The tutorial subset `tutorial_binary` is a binary task with label column `label`.

## Notes

- The H1 ATAC-seq region files require a Synapse account and are not included.

## Source

If you use this dataset, please cite:

- Pampari A, Shcherbina A, Kvon E, et al. ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility reveal cis-regulatory sequence syntax, transcription factor footprints and regulatory variants. bioRxiv 2024. doi:10.1101/2024.12.25.630221

## License

Declared license: CC-BY-4.0. The terms of the original data providers listed in Source also apply.