---
title: C-Origami
canonical_url: "https://www.modelscope.cn/datasets/User040218/C-Origami"
md_url: "https://www.modelscope.cn/datasets/User040218/C-Origami.md"
repository: User040218/C-Origami
last_updated: 2026-11-01
license: cc-by-4.0
storage_size: "34 GB"
downloads: 14218
stars: 0
---

# C-Origami

> C-Origami - User040218 在 ModelScope 开源的数据集。C-Origami GM12878 and IMR-90 numeric training data with Hi-C contact-matrix targets; the web preview shows a row index.

User040218/C-Origami 是 ModelScope 魔搭社区上的数据集，存储大小 34 GB，采用 cc-by-4.0 许可。

- **Repository**: User040218/C-Origami
- **License**: cc-by-4.0
- **Storage size**: 34 GB
- **Downloads**: 14218
- **Stars**: 0
- **Last updated**: 2026-11-01

Source: https://www.modelscope.cn/datasets/User040218/C-Origami

---

# C-Origami

**Numeric training tensors for 3D chromatin organization prediction in GM12878 and IMR-90** — regression.

C-Origami predicts cell-type-specific 3D chromatin organization from DNA sequence, CTCF ChIP-seq and ATAC-seq. The authors' Zenodo record provides the datasets generated from the IMR-90 and GM12878 cell lines together with the pretrained model weights.

This release keeps the original numeric training tensors: inputs are stored as `inputs_embeds` and targets are 256 × 256 Hi-C contact matrices (65,536 values), with the authors' train/validation/test partitions. The previewable subsets `index_gm12878` and `index_imr90` are row-level indices (10,356 rows) recording coordinates and the location of each row in the native files; use the code below to load the training tensors.

Official dataset description: [Zenodo 7226561](https://zenodo.org/records/7226561). See also: [C.Origami (GitHub)](https://github.com/tanjimin/C.Origami).

## Subsets (2)

| Subset | Splits (rows) | Description |
|---|---|---|
| `index_gm12878` | test 178 · train 4,757 · validation 243 | GM12878 row index |
| `index_imr90` | test 178 · train 4,757 · validation 243 | IMR-90 row index |

10,356 rows in total.

## Format

Parquet. Main columns:

| Column | Type | Description |
|---|---|---|
| `chromosome` / `start` / `end` | string / int64 | Sample window coordinates (index) |
| `source_file` / `source_row` | string / int64 | Native file and row (index) |
| `inputs_embeds_shape` / `labels_shape` | string | Full shapes of the native inputs and targets (index) |

## Usage

Example with [DNALLM](https://github.com/zhangtaolab/DNALLM): subset `gm12878`, regression, `num_labels=65536`.

```python
import json
from modelscope import snapshot_download
from dnallm import DNADataset

repo = "User040218/C-Origami"
root = snapshot_download(repo_id=repo, repo_type="dataset", allow_patterns=["source/native-file-map.json"])
files = json.load(open(f"{root}/source/native-file-map.json", encoding="utf-8"))["gm12878"]
root = snapshot_download(repo_id=repo, repo_type="dataset",
                         allow_patterns=[f["path"] for split in files.values() for f in split])
data = DNADataset.load_local_data(
    {split: [f"{root}/{f['path']}" for f in items] for split, items in files.items()},
    seq_col="sequence", label_col="labels")
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`.

The index subsets shown in the web preview are for browsing and locating rows only; the code above downloads and reads the native files listed in `source/native-file-map.json`.

## Notes

- The index columns `inputs_embeds_file_min`, `inputs_embeds_file_max`, `labels_file_min` and `labels_file_max` are empty (null) in this release and can be ignored.

## Source

If you use this dataset, please cite:

- Tan J, Shenker-Tauris N, Rodriguez-Hernaez J, et al. Cell-type-specific prediction of 3D chromatin organization enables high-throughput in silico genetic screening. Nature Biotechnology 2023;41(8):1140-1150. doi:10.1038/s41587-022-01612-8

## License

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