---
title: EPCOT
canonical_url: "https://www.modelscope.cn/datasets/User040218/EPCOT"
md_url: "https://www.modelscope.cn/datasets/User040218/EPCOT.md"
repository: User040218/EPCOT
last_updated: 2026-11-01
license: cc-by-4.0
storage_size: "29 GB"
downloads: 981
stars: 0
---

# EPCOT

> EPCOT - User040218 在 ModelScope 开源的数据集。EPCOT pre-training data: 1,600 bp sequences, DNase inputs and multilabel epigenomic targets for four cell lines (hg38).

User040218/EPCOT 是 ModelScope 魔搭社区上的数据集，存储大小 29 GB，采用 cc-by-4.0 许可。

- **Repository**: User040218/EPCOT
- **License**: cc-by-4.0
- **Storage size**: 29 GB
- **Downloads**: 981
- **Stars**: 0
- **Last updated**: 2026-11-01

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

---

# EPCOT

**EPCOT pre-training data: 1,600 bp sequences, DNase inputs and multilabel epigenomic targets for four cell lines (hg38)** — multilabel classification.

EPCOT is a generalizable framework for predicting the epigenome, chromatin organization and transcriptome. Its Zenodo record contains the trained models, processed DNase-seq/ATAC-seq data, the pre-training targets (binary epigenomic feature activity) and the accession numbers of the ChIP-seq data used for pre-training.

This release covers the pre-training data of four cell lines (hg38): each row is a 1 kb target bin; `sequence` is the 1,600 bp sequence including 300 bp on each side, `inputs_embeds` holds the DNase input and `labels` the cell-line-specific multilabel targets.

Official dataset description: [Zenodo 7485616](https://zenodo.org/records/7485616). See also: [EPCOT (GitHub)](https://github.com/zzh24zzh/EPCOT).

## Subsets (4)

| Subset | Splits (rows) | Description |
|---|---|---|
| `GM12878` | train 2,874,990 | 125 labels |
| `HepG2` | train 2,874,990 | 208 labels |
| `K562` | train 2,874,990 | 221 labels |
| `MCF-7` | train 2,874,990 | 103 labels |

11,499,960 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `sequence` | string | 1,600 bp DNA sequence (hg38) |
| `inputs_embeds` | list<list<float32>>[11] | DNase numeric input |
| `labels` | list<float32>[125] | Multilabel epigenomic targets |
| `chromosome` / `start` / `end` | string / int64 | 1 kb target bin |

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

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `MCF-7`, multilabel classification, `num_labels=103`.

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

ds = MsDataset.load("User040218/EPCOT", subset_name="MCF-7")
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

- Only the pre-training data are included; the downstream Hi-C, Micro-C and RNA task data are not.
- Each cell line has a single train split.

## Source

If you use this dataset, please cite:

- Zhang Z, Feng F, Qiu Y, Liu J. A generalizable framework to comprehensively predict epigenome, chromatin organization, and transcriptome. Nucleic Acids Research 2023;51(12):5931-5947. doi:10.1093/nar/gkad436

## License

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