# CDT2-data

Data files for [Central Dogma Transformer II (CDT-II)](https://github.com/nobusama/CDT2).

## Files

| File | Description | Size |
|------|-------------|------|
| `morris_celllevel_effects_2361.h5` | Cell-level perturbation effects (TSS, 2,361 genes) | 41 MB |
| `morris_snp_celllevel_effects_2361.h5` | Cell-level perturbation effects (SNP, 2,361 genes) | 34 MB |
| `k562_gene_embeddings_aligned.h5` | Gene embeddings from scGPT (2,360 genes) | 4.4 MB |
| `cdt_morris_celllevel_best.pt` | Trained CDT-II model weights | 80 MB |

## Enformer Embeddings (Not Included)

The Enformer embeddings must be generated using the notebooks in the CDT2 repository:

- `morris_28genes_enformer.h5` - Generate with `notebooks/embeddings/Morris_28genes_Enformer.ipynb`
- `morris_snp_enformer.h5` - Generate with `notebooks/embeddings/Morris_SNP_Enformer.ipynb`

These notebooks run Enformer inference on Google Colab and save the embeddings to Google Drive.

## Usage

```python
from huggingface_hub import hf_hub_download

# Download cell-level effects
effects_path = hf_hub_download(
    repo_id="nobusama17/CDT2-data",
    filename="morris_celllevel_effects_2361.h5",
    repo_type="dataset"
)

# Download model weights
model_path = hf_hub_download(
    repo_id="nobusama17/CDT2-data",
    filename="cdt_morris_celllevel_best.pt",
    repo_type="dataset"
)
```

## Data Source

- Morris STING-seq v2 dataset (60,505 K562 cells, 447 perturbation loci)
- Reference: [Morris et al. Science 2023](https://www.science.org/doi/10.1126/science.adh7699)

## Citation

```bibtex
@article{ota2025cdtii,
  title={Central Dogma Transformer II: An AI Microscope for Understanding Cellular Regulatory Mechanisms},
  author={Ota, Nobuyuki},
  journal={bioRxiv},
  year={2025}
}
```

## License

MIT License