---
title: Basset
canonical_url: "https://www.modelscope.cn/datasets/User040218/Basset"
md_url: "https://www.modelscope.cn/datasets/User040218/Basset.md"
repository: User040218/Basset
last_updated: 2026-11-01
license: mit
storage_size: "3.6 GB"
downloads: 402
stars: 0
---

# Basset

> Basset - User040218 在 ModelScope 开源的数据集。Basset: 600 bp DNA sequences with accessibility labels for 164 DNase-seq samples (multilabel classification).

User040218/Basset 是 ModelScope 魔搭社区上的数据集，存储大小 3.6 GB，采用 mit 许可。

- **Repository**: User040218/Basset
- **License**: mit
- **Storage size**: 3.6 GB
- **Downloads**: 402
- **Stars**: 0
- **Last updated**: 2026-11-01

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

---

# Basset

**DNA sequences with chromatin accessibility labels across 164 DNase-seq samples** — multilabel classification.

Basset learns the regulatory code of the accessible genome with deep convolutional neural networks. Its training data are 600 bp hg19 sequences from merged DNase I hypersensitivity peaks of 164 ENCODE and Roadmap Epigenomics samples; each sequence carries 164 binary labels marking the samples in which the region is accessible.

This release keeps the authors' intervals, activity table and partitions: 1,880,000 training, 70,000 validation and 71,886 test sequences.

Links: [Basset (GitHub)](https://github.com/davek44/Basset).

## Splits

Subset: `default`

| Split | Rows |
|---|---:|
| train | 1,880,000 |
| validation | 70,000 |
| test | 71,886 |

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `sequence` | string | 600 bp DNA sequence (hg19) |
| `labels` | list<float32>[164] | 164 binary accessibility labels |
| `chromosome` / `start` / `end` / `strand` | string | Original BED interval |
| `active_indices` | string | Indices of the samples in which the region is accessible (original text) |

The other 3 columns are original source fields kept under their original names.

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `default`, multilabel classification, `num_labels=164`.

```python
from modelscope.msdatasets import MsDataset
from dnallm import DNADataset

ds = MsDataset.load("User040218/Basset", subset_name="default")
# tokenizer: the tokenizer of your DNALLM model; set max_length to match its tokenization
data = DNADataset(ds, tokenizer=tokenizer, max_length=602)
```

## Notes

- The paper reports 2,071,886 sites (1,930,000 for training); the authors' currently released BED/activity files contain 2,021,886 sites. All released rows are kept; the source of the 50,000-site difference is unresolved.
- `start` and `end` are stored as text, as in the source; cast them to integers before arithmetic.

## Source

If you use this dataset, please cite:

- Kelley DR, Snoek J, Rinn JL. Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks. Genome Research 2016;26(7):990-999. doi:10.1101/gr.200535.115

## License

Declared license: MIT. The terms of the original data providers listed in Source also apply.