---
title: Reformer
canonical_url: "https://www.modelscope.cn/datasets/User040218/Reformer"
md_url: "https://www.modelscope.cn/datasets/User040218/Reformer.md"
repository: User040218/Reformer
last_updated: 2026-11-01
license: cc-by-4.0
storage_size: "1.8 GB"
downloads: 6590
stars: 0
---

# Reformer

> Reformer - User040218 在 ModelScope 开源的数据集。Reformer ENCODE eCLIP data: 223 RBP × cell-line binding-site classification subsets plus single-base binding profiles.

User040218/Reformer 是 ModelScope 魔搭社区上的数据集，存储大小 1.8 GB，采用 cc-by-4.0 许可。

- **Repository**: User040218/Reformer
- **License**: cc-by-4.0
- **Storage size**: 1.8 GB
- **Downloads**: 6590
- **Stars**: 0
- **Last updated**: 2026-11-01

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

---

# Reformer

**ENCODE eCLIP data for Reformer: 223 RBP × cell-line binding-site classification subsets and a single-base binding-profile set**.

The authors' Zenodo record provides two training datasets from ENCODE eCLIP-seq experiments: `encode_eclip_bc.h5` for Reformer-BC, which distinguishes binding from non-binding sites, and `encode_eclip.h5` for Reformer, which predicts binding affinity at single-base resolution.

This release splits the classification data by RBP and cell line into 223 subsets, each with train/validation/test splits, and stores the binding-profile data in the `profile` subset (512 nt sequences with per-base binding profiles); 2,958,884 rows in total.

Official dataset description: [Zenodo 14027315](https://zenodo.org/records/14027315). See also: [Reformer (GitHub)](https://github.com/xilinshen/Reformer).

## Subsets (224)

The 223 classification subsets are named `<RBP>_<cell>` (for example `AARS_K562`): 118 K562, 103 HepG2 and 2 adrenal gland. The `profile` subset holds the binding-profile data.

224 subsets, 2,958,884 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `sequence` | string | RNA sequence |
| `label` | int64 | Binding site (0/1, classification subsets) |
| `name` | string | Region name |

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `AARS_K562`, binary classification, `num_labels=2`.

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

ds = MsDataset.load("User040218/Reformer", subset_name="AARS_K562")
ds = ds.rename_column("label", "labels")
# tokenizer: the tokenizer of your DNALLM model; set max_length to match its tokenization
data = DNADataset(ds, tokenizer=tokenizer, max_length=512)
```

## Notes

- Inputs of the `profile` subset are the authors' token IDs (including the condition barcode); keep the authors' encoding.

## Source

If you use this dataset, please cite:

- Shen X, Hou Y, Wang X, et al. A deep learning model for characterizing protein-RNA interactions from sequences at single-base resolution. Patterns 2025;6(1):101150. doi:10.1016/j.patter.2024.101150

## License

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