---
title: ProCapNet
canonical_url: "https://www.modelscope.cn/datasets/User040218/ProCapNet"
md_url: "https://www.modelscope.cn/datasets/User040218/ProCapNet.md"
repository: User040218/ProCapNet
last_updated: 2026-11-01
license: cc-by-4.0
storage_size: "1.8 GB"
downloads: 853
stars: 0
---

# ProCapNet

> ProCapNet - User040218 在 ModelScope 开源的数据集。ProCapNet: 2,114 bp sequences with stranded base-resolution PRO-cap initiation profiles in six cell lines.

User040218/ProCapNet 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/ProCapNet
- **License**: cc-by-4.0
- **Storage size**: 1.8 GB
- **Downloads**: 853
- **Stars**: 0
- **Last updated**: 2026-11-01

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

---

# ProCapNet

**2,114 bp sequences with stranded base-resolution PRO-cap transcription initiation profiles in six cell lines** — regression.

ProCapNet predicts base-resolution, stranded PRO-cap transcription initiation signal from DNA sequence. This release contains the six cell lines of the paper (A673, Caco-2, Calu3, HUVEC, K562, MCF 10A), one subset per cell line with train/validation/test splits.

Each example is a 2,114 bp GRCh38 sequence; `labels` holds the plus-strand and minus-strand counts over the central 1,000 bp (2,000 values), `mask` the matching k36 mappability mask, and `fold_1`–`fold_7` all author fold assignments. 870,645 rows in total.

Links: [ProCapNet (GitHub)](https://github.com/kundajelab/ProCapNet).

## Subsets (6)

| Subset | Splits (rows) |
|---|---|
| `A673` | train 49,076 · validation 9,056 · test 9,073 |
| `CACO2` | train 94,035 · validation 17,913 · test 17,357 |
| `CALU3` | train 86,322 · validation 15,654 · test 16,624 |
| `HUVEC` | train 157,552 · validation 31,192 · test 32,491 |
| `K562` | train 82,200 · validation 14,477 · test 16,215 |
| `MCF10A` | train 156,774 · validation 30,671 · test 33,963 |

870,645 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `sequence` | string | 2,114 bp DNA sequence (GRCh38) |
| `labels` | list<float32>[2000] | Per-base counts, 1,000 plus-strand + 1,000 minus-strand |
| `mask` | list<int8>[1000] | Mappability mask |
| `region_type` / `chrom` / `start` / `end` | string / int64 | Region type and coordinates |
| `fold_1 … fold_7` | string | Author fold assignments |

The other 9 columns are original source fields kept under their original names.

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `HUVEC`, regression, `num_labels=2000`.

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

ds = MsDataset.load("User040218/ProCapNet", subset_name="HUVEC")
# tokenizer: the tokenizer of your DNALLM model; set max_length to match its tokenization
data = DNADataset(ds, tokenizer=tokenizer, max_length=2116)
```

## Source

If you use this dataset, please cite:

- Cochran K, Yin M, Mantripragada A, et al. Dissecting the cis-regulatory syntax of transcription initiation with deep learning. bioRxiv 2024. doi:10.1101/2024.05.28.596138

## License

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