---
title: gkm-align
canonical_url: "https://www.modelscope.cn/datasets/User040218/gkm-align"
md_url: "https://www.modelscope.cn/datasets/User040218/gkm-align.md"
repository: User040218/gkm-align
last_updated: 2026-11-01
license: other
storage_size: "3.9 GB"
downloads: 1205
stars: 0
---

# gkm-align

> gkm-align - User040218 在 ModelScope 开源的数据集。gkm-align human/mouse DHS, enhancer, mapping and conserved-enhancer data; four subsets support sequence regression.

User040218/gkm-align 是 ModelScope 魔搭社区上的数据集，存储大小 3.9 GB，采用 other 许可。

- **Repository**: User040218/gkm-align
- **License**: other
- **Storage size**: 3.9 GB
- **Downloads**: 1205
- **Stars**: 0
- **Last updated**: 2026-11-01

Source: https://www.modelscope.cn/datasets/User040218/gkm-align

---

# gkm-align

**Human/mouse DHS, enhancer, alignment-mapping and conserved-enhancer data from the gkm-align paper; four subsets support sequence regression**.

gkm-align is a whole-genome alignment algorithm that maps distal enhancers conserved between distant mammals, such as human and mouse, by finding alignment paths with maximal gapped-k-mer compositional similarity. The authors' project page provides the paper's supplementary data, including DHS regions, enhancer sets and mapping results.

This release includes seven of these resources: `dhs_human`, `dhs_mouse`, `enhancers_human` and `enhancers_mouse` contain hg38/mm10 sequences with gkm-SVM scores for sequence regression (label column `score`); `mapping_weighted`, `mapping_unweighted` and `conserved` contain alignment mappings and conserved enhancers as auxiliary data.

Official dataset description: [gkm-align project page](https://beerlab.org/gkmalign/). See also: [gkm-align (GitHub)](https://github.com/oh-jinwoo94/gkm-align).

## Subsets (7)

| Subset | Splits (rows) | Description |
|---|---|---|
| `conserved` | train 5,856,148 | Auxiliary |
| `dhs_human` | train 3,532,505 | Sequence regression |
| `dhs_mouse` | train 3,451,467 | Sequence regression |
| `enhancers_human` | train 1,526,300 | Sequence regression |
| `enhancers_mouse` | train 1,850,633 | Sequence regression |
| `mapping_unweighted` | train 1,236,288 | Auxiliary |
| `mapping_weighted` | train 1,401,734 | Auxiliary |

18,855,075 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `sequence` | string | Region DNA sequence (hg38/mm10) |
| `score` | float64 | gkm-SVM score |
| `chrom` / `start` / `end` | string / int64 | Region coordinates |
| `biosample` / `encode_model` | string | Biosample and ENCODE model |

The other 7 columns are original source fields kept under their original names.

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `dhs_mouse`, regression, `num_labels=1`.

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

ds = MsDataset.load("User040218/gkm-align", subset_name="dhs_mouse")
ds = ds.rename_column("score", "labels")
# tokenizer: the tokenizer of your DNALLM model; set max_length to match its tokenization
data = DNADataset(ds, tokenizer=tokenizer, max_length=302)
```

## Notes

- The negative training sequences were not released by the authors and are not synthesized here.
- Each subset has a single train split.

## Source

If you use this dataset, please cite:

- Oh JW, Beer MA. Gapped-kmer sequence modeling robustly identifies regulatory vocabularies and distal enhancers conserved between evolutionarily distant mammals. Nature Communications 2024;15(1):6464. doi:10.1038/s41467-024-50708-z

## License

Declared as `other`: use of the data follows the terms of the original data providers listed in Source.