Dataset Card for LucaVirus-OpenVirus-Prot

1. Dataset Summary

LucaVirus-OpenVirus-Prot is a large-scale proteomics dataset consisting exclusively of viral protein sequences. It is a specialized subset of the OpenVirus corpus, specifically curated for the pre-training of the LucaVirus-Prot (or LucaVirus-Protein) foundation model.

This dataset provides a comprehensive representation of the viral proteosphere, comprising 5.2 million protein sequences. It is designed to enable biological models to learn the "language of proteins," capturing structural motifs, functional domains, and evolutionary signatures across a vast array of viral families.

2. Dataset Statistics

The dataset focuses strictly on amino acid sequences:

Feature	Count / Description
Total Sequences	5.2 Million
Sequence Type	Protein (Amino Acids)
obj_type Identifier	prot (Exclusive)
Primary Use	Pre-training for LucaVirus-Prot

3. Data Structure & Format

3.1 File Organization

The dataset is distributed as a compressed .tar archive. Upon extraction, the data is partitioned into three standard machine-learning subsets:

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

Each directory (train, dev, test) contains one or more CSV files with headers.

3.2 CSV Schema

All CSV files follow a consistent four-column schema:

Column Name	Description	Details
obj_id	Sample ID	Unique identifier for each protein sequence.
obj_type	Sequence Type	Set to prot for all entries in this dataset.
obj_seq	Sequence Content	Raw amino acid string (standard IUPAC codes).
obj_label	Annotations	Metadata, taxonomic info, or functional labels associated with the protein (Annotation, Bio Knowledge)

4. Intended Use

• Protein Foundation Modeling: Building models like LucaVirus-Prot that specialize in understanding protein sequences and their biophysical properties.
• Functional Annotation: Developing tools to predict viral protein functions, domains, and active sites.
• Virus-Host Interaction: Studying how viral proteins interact with host cellular machinery based on sequence patterns.

5. Usage Example

You can extract the archive and load the protein data using the following Python snippet:

import tarfile
import pandas as pd
import os

# 1. Extract the protein dataset
with tarfile.open("LucaVirus-OpenVirus-Prot.tar.gz", "r:gz") as tar:
    tar.extractall(path="./LucaVirus-OpenVirus-Prot")

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

# 2. Load a sample from the training set
train_path = "./LucaVirus-OpenVirus-Prot/dataset/v1.0/train"
csv_files = [f for f in os.listdir(train_path) if f.endswith('.csv')]

if csv_files:
    # Load the first CSV file
    df = pd.read_csv(os.path.join(train_path, csv_files[0]))
    
    # Verify the sequence type
    print(f"Loaded {len(df)} protein sequences.")
    print(df[['obj_id', 'obj_seq', 'obj_label']].head())

6. Related Resources

This dataset is a core component of the LucaGroup biological modeling ecosystem.

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

7. Citation

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

@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.