yangyz1230/space

This dataset card contains data from the original Basenji project. The original Basenji dataset has two main limitations:

1. Format: Data is stored in TensorFlow format, which is not directly compatible with PyTorch workflows
2. Cost: Users need to pay Google Cloud storage fees to download the data

To facilitate PyTorch-based training, we have downloaded and converted the data to H5 format. With permission from the original Basenji authors, we are releasing the H5-formatted data here for free access.

📁 Key Files

• human_train.h5, human_valid.h5, human_test.h5
• mouse_train.h5, mouse_valid.h5, mouse_test.h5

📦 File Splitting & Reconstruction

Since the training files exceed 50GB and cannot be directly uploaded to 🤗 Hugging Face, we split them using the following commands:

split -b 45G -d -a 2 human_train.h5 human_train_part_
split -b 45G -d -a 2 mouse_train.h5 mouse_train_part_

After downloading all part files, you need to reconstruct the original H5 files:

# Reconstruct human_train.h5
cat human_train_part_* > human_train.h5

# Reconstruct mouse_train.h5  
cat mouse_train_part_* > mouse_train.h5

📖 Citation

If you find this dataset useful, please cite both the original Basenji paper and our work:

@article{kelley2018sequential,
  title={Sequential regulatory activity prediction across chromosomes with convolutional neural networks},
  author={Kelley, David R and Reshef, Yakir A and Bileschi, Maxwell and Belanger, David and McLean, Cory Y and Snoek, Jasper},
  journal={Genome research},
  volume={28},
  number={5},
  pages={739--750},
  year={2018},
  publisher={Cold Spring Harbor Lab}
}

@misc{yang2025spacegenomicprofilepredictor,
  title={SPACE: Your Genomic Profile Predictor is a Powerful DNA Foundation Model}, 
  author={Zhao Yang and Jiwei Zhu and Bing Su},
  year={2025},
  eprint={2506.01833},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2506.01833}
}