Dataset Card for LucaVirus-OpenVirus-Gene-Prot

1. Dataset Summary

LucaVirus-OpenVirus-Gene-Prot is the complete, multi-modal OpenVirus corpus, curated for the pre-training of the LucaVirus biological foundation model. This dataset provides a massive-scale collection of viral sequences, bridging the gap between genomic (nucleotide) and proteomic (protein) data.

The corpus comprises 15.7 million(10.4M nucleotide sequences and 5.2M protein sequences) non-redundant viral sequences, providing a robust foundation for learning the complex language of viral evolution and the "central dogma" of viral biology.

2. Dataset Statistics

Data Type	Count	obj_type Identifier
Nucleotide (Genomes)	10.4 Million	gene
Protein (Amino Acids)	5.2 Million	prot
Total Sequences	15.7 Million	-

3. Data Structure

The dataset is provided as a compressed .tar archive. Once extracted, the directory structure follows a standard machine-learning split:

LucaVirus-OpenVirus-Gene-Prot/dataset/v1.0/
├── train/          # Training set (primary corpus for pre-training)
├── dev/            # Validation set (for hyperparameter tuning)
└── test/           # Test set (for final evaluation)

Each directory contains one or more CSV files with headers.

Data Schema

Each CSV file includes the following columns:

Column Name	Description	Details
obj_id	Sample ID	Unique identifier for the sample.
obj_type	Sequence Type	Sequence modality: gene (nucleotide) or prot (protein).
obj_seq	Sequence Content	The raw biological sequence (AT(U)GCN for gene; Amino Acids for prot).
obj_label	Label	Metadata, taxonomic info, or functional labels associated with the genome and proteins (Annotation, Bio Knowledge)

4. Dataset Intent

This dataset is specifically designed for:

• Foundation Model Pre-training: Building models that can process both DNA/RNA and Protein sequences.
• Cross-modal Learning: Understanding the translation and structural relationships within viral biology.
• Viral Research: A large-scale benchmark for viral sequence classification, functional annotation, and mutation analysis.

5. Usage

Loading with Python

You can use standard Python libraries to process the data:

import pandas as pd
import tarfile
import os

# Example: Extracting and reading a file
with tarfile.open("LucaVirus-OpenVirus-Gene-Prot.tar.gz", "r:gz") as tar:
    tar.extractall(path="./LucaVirus-OpenVirus-Gene-Prot/")

with tarfile.open("./LucaVirus-OpenVirus-Gene-Prot/dataset.tar.gz", "r:gz") as tar:
    tar.extractall(path="./LucaVirus-OpenVirus-Gene-Prot/dataset/")

# Read a specific CSV from the train set
df = pd.read_csv("../LucaVirus-OpenVirus-Gene-Prot/dataset/v1.0/train/3072_train_1.csv")
print(df.head())

6. Pre-training with LucaVirus

This dataset is the primary source for the LucaVirus model family.

• Full Corpus (Gene + Prot): LucaVirus-OpenVirus-Gene
• Protein Subset: LucaVirus-OpenVirus-Prot
• Models: Visit the LucaVirus Collection.

7. Citation

If you use this dataset in your research, please cite the following:

@article{lucavirus2025,
  title={Predicting the Evolutionary and Functional Landscapes of Viruses with a Unified Nucleotide-Protein Language Model: LucaVirus.},
  author={Pan, Yuan-Fei* and He, Yong*. et al.},
  journal={bioRxiv},
  year={2025},
  url={https://www.biorxiv.org/content/early/2025/06/20/2025.06.14.659722}
}

8. License

This dataset is released under the MIT License.

9. Contact

For further information, please visit the LucaGroup GitHub, email to: [YongHe: [email protected], [email protected]], or contact the team via the Hugging Face organization profile.