---
title: Loop-Extrusion-Model
canonical_url: "https://www.modelscope.cn/datasets/User040218/Loop-Extrusion-Model"
md_url: "https://www.modelscope.cn/datasets/User040218/Loop-Extrusion-Model.md"
repository: User040218/Loop-Extrusion-Model
last_updated: 2026-11-01
license: other
storage_size: "103 MB"
downloads: 191
stars: 0
---

# Loop-Extrusion-Model

> Loop-Extrusion-Model - User040218 在 ModelScope 开源的数据集。Loop competition and extrusion model: CTCF peak-pair tables (GM12878, HeLa) with 7 features (binary classification).

User040218/Loop-Extrusion-Model 是 ModelScope 魔搭社区上的数据集，存储大小 103 MB，采用 other 许可。

- **Repository**: User040218/Loop-Extrusion-Model
- **License**: other
- **Storage size**: 103 MB
- **Downloads**: 191
- **Stars**: 0
- **Last updated**: 2026-11-01

Source: https://www.modelscope.cn/datasets/User040218/Loop-Extrusion-Model

---

# Loop-Extrusion-Model

**CTCF peak-pair training tables (GM12878, HeLa) for the loop competition and extrusion model, with 7 features** — binary classification.

This mathematical model predicts CTCF interaction specificity from loop competition and extrusion. The authors' GitHub repository provides example training inputs covering all CTCF peaks and peak pairs within 1 Mb on GM12878 chromosome 1.

This release includes the complete peak-pair tables for GM12878 and HeLa, 1,932,776 trainable examples, keeping the original columns (peak positions, heights, distance, orientations and motif scores); the 7 features are packed into `inputs_embeds` and `labels` marks an interaction.

Official dataset description: [Loop-Extrusion-Model (GitHub)](https://github.com/wangxi001/Loop-Extrusion-Model).

## Subsets (2)

| Subset | Splits (rows) |
|---|---|
| `GM12878` | train 1,119,764 |
| `HeLa` | train 813,012 |

1,932,776 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `peak1` / `peak2` | int64 | Positions of the two CTCF peaks |
| `height1` / `height2` / `distance` / `strand1` / `strand2` / `motif1` / `motif2` | float64 / int64 | Original features |
| `labels` | int64 | Interaction (0/1) |
| `inputs_embeds` | list<list<float64>>[1] | Numeric input built from the 7 features |

The other 5 columns are original source fields kept under their original names.

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `HeLa`, binary classification, `num_labels=2`.

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

ds = MsDataset.load("User040218/Loop-Extrusion-Model", subset_name="HeLa")
data = DNADataset(ds)
data.dataset.set_format("torch", columns=["inputs_embeds", "attention_mask", "labels"])
```

Inputs are numeric features in `inputs_embeds`; no tokenization is needed. Use a model that accepts `inputs_embeds`.

## Notes

- Examples follow the authors' evaluation rule (label ≥ 0 and both motif scores > 0); excluded rows remain in the original `pairs/` tables.
- There is no train/test split.

## Source

If you use this dataset, please cite:

- Xi W, Beer MA. Loop competition and extrusion model predicts CTCF interaction specificity. Nature Communications 2021;12(1):1046. doi:10.1038/s41467-021-21368-0

## License

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