duttaprat/HVUE

---
language:
- en
license: cc-by-4.0
task_categories:
- text-classification
tags:
- biology
- virology
- genomics
- pathogenicity
- benchmark
- viral-genomics
size_categories:
- 10K<n<100K
---

# HVUE: Human Virome Understanding Evaluation

## Dataset Description

HVUE (Human Virome Understanding Evaluation) is a comprehensive benchmark for evaluating foundation models on viral genomics tasks. The benchmark comprises 7 curated datasets across 3 epidemiologically critical prediction tasks:

- **Pathogenicity Classification** (3 datasets)
- **Host Tropism Prediction** (1 dataset)
- **Transmissibility Assessment** (3 datasets)

**Paper**: *HViLM: A Foundation Model for Viral Genomics Enables Multi-Task Prediction of Pathogenicity, Transmissibility, and Host Tropism*

**Authors**: Pratik Dutta, Jack Vaska, Pallavi Surana, Rekha Sathian, Max Chao, Zhihan Zhou, Han Liu, and Ramana V. Davuluri

**GitHub**: https://github.com/duttaprat/HViLM

## Dataset Structure

### Pathogenicity Classification

**CINI Dataset**
- 159 sequences across 4 viral families
- Manual literature-based curation
- Binary classification: pathogenic vs non-pathogenic

**BVBRC-CoV Dataset**
- 18,066 coronavirus sequences
- Distinguishes human-pathogenic (SARS-CoV-2, MERS-CoV, etc.) from animal-restricted strains

**BVBRC-Calici Dataset**
- 31,089 calicivirus sequences
- Clinical evidence and isolation source-based labels

### Host Tropism Prediction

**VHDB Dataset**
- 9,428 sequences spanning 30 viral families
- Binary classification: human-tropic (13.1%) vs non-human-tropic (86.9%)
- Experimentally validated host range annotations

### Transmissibility Prediction

**Coronaviridae Dataset**
- ~3,000 coronavirus sequences
- R₀-based classification: R₀<1 vs R₀≥1

**Orthomyxoviridae Dataset**
- ~2,500 influenza sequences
- R₀-based classification

**Caliciviridae Dataset**
- ~1,800 calicivirus sequences
- R₀-based classification

## Data Format

Each dataset contains three splits:
- `train.csv`
- `dev.csv`
- `test.csv`

CSV columns:
- `sequence`: Viral genomic sequence (250-1000 bp)
- `label`: Binary label (0 or 1)

## Usage
```python
from datasets import load_dataset

# Load entire benchmark
hvue = load_dataset("duttaprat/HVUE")

# Load specific task
patho_cini = load_dataset("duttaprat/HVUE", data_files="pathogenicity/CINI/*.csv")

# Load specific split
train_data = load_dataset("duttaprat/HVUE", data_files="pathogenicity/CINI/train.csv")
```

## Citation
```bibtex
@article{dutta2025hvilm,
title={HViLM: A Foundation Model for Viral Genomics Enables Multi-Task Prediction of Pathogenicity, Transmissibility, and Host Tropism},
author={Dutta, Pratik and Vaska, Jack and Surana, Pallavi and Sathian, Rekha and Chao, Max and Zhou, Zhihan and Liu, Han and Davuluri, Ramana V.},
journal={Submitted to RECOMB},
year={2025}
}
```

## License

CC-BY-4.0

## Contact

- Pratik Dutta: [e-mail]
- GitHub Issues: https://github.com/duttaprat/HViLM/issues