README.md

---
license: cc-by-4.0
task_categories:
- text-classification
tags:
- public health
- bioinformatics
- virus
- proteomics
- vaccine development
- antigen
- machine learning
- reverse vaccinology
- viral proteins
- protegen
- uniprot
pretty_name: VPAgs-Dataset4ML
size_categories:
- 1K<n<10K
---
# Dataset Card for VPAgs-Dataset4ML

## Dataset Details

### Dataset Description

**VPAgs-Dataset4ML** comprises 2,145 viral protein sequences, curated to facilitate the development of machine learning models capable of predicting viral protective antigens (PAgs). These antigens are crucial for designing vaccines against various viral pathogens. The dataset is divided into two categories: 210 protective antigens (positive class) and 1,935 non-protective protein sequences (negative class), derived from the Protegen database and UniProt, respectively. This collection aims to support and accelerate research in reverse vaccinology, providing a valuable resource for bioinformatics and public health.

- **Curated by:** Zakia Salod and Ozayr Mahomed from the University of KwaZulu-Natal.
- **Funded by** National Research Foundation (NRF) of South Africa (grant number 130187) and College of Health Sciences (CHS) of the University of KwaZulu-Natal (UKZN) in Durban, Kwa-Zulu-Natal, South Africa.
- **Language(s) (NLP):** English.
- **License:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)

### Dataset Sources

- **Repository:** Mendeley Data - [VPAgs-Dataset4ML](https://doi.org/10.17632/w78tyrjz4z.1)
- **Paper** Salod, Z.; Mahomed, O. VPAgs-Dataset4ML: A Dataset to Predict Viral Protective Antigens for Machine Learning-Based Reverse Vaccinology. Data 2023, 8, 41. [https://doi.org/10.3390/data8020041](https://doi.org/10.3390/data8020041).

## Uses

### Direct Use

This dataset serves as an invaluable asset for developing and testing machine learning algorithms aimed at identifying potential vaccine candidates. Its application extends beyond academic research, offering insights that could significantly impact vaccine development strategies, particularly in the realm of emerging viral threats.

## Dataset Structure

### Data Instances

```
{
    "sequence": "MATLLRSLALFKRNKDKPPITSGSGGAIRGIKHIIIVPIPGDSSITTRSRLLDRLVRLIGNPDVSGPKLTGALIGILSLFVESPGQLIQRITDDPDVSIRLLEVVQSDQSQSGLTFASRGTNMEDEADQYFSHDDPSSSDQSRSGWFENKEISDIEVQDPEGFNMILGTILAQIWVLLAKAVTAPDTAADSELRRWIKYTQQRRVVGEFRLERKWLDVVRNRIAEDLSLRRFMVALILDIKRTPGNKPRIAEMICDIDTYIVEAGLASFILTIKFGIETMYPALGLHEFAGELSTLESLMNLYQQMGETAPYMVILENSIQNKFSAGSYPLLWSYAMGVGVELENSMGGLNFGRSYFDPAYFRLGQEMVRRSAGKVSSTLASELGITAEDARLVSEIAMHTTEDRISRAVGPRQAQVSFLHGDQSENELPGLGGKEDRRVKQGRGEARESYRETGSSRASDARAAHPPTSMPLDIDTASESGQDPQDSRRSADALLRLQAMAGILEEQGSDTDTPRVYNDRDLLD",
    "label": "1"
}
```

### Data Fields

- `sequence`: A string representing the amino acid sequence of a viral protein.
- `label`: An integer indicating whether the sequence is a protective antigen (1) or not (0).

### Data Splits

The dataset has not been split into training and testing sets, to allow for flexibility.
You may split the dataset into training and testing sets, based on your preferred ratio.

## Dataset Creation

### Curation Rationale

The dataset was curated to address the need for a machine learning-ready dataset containing labeled protective (positive) and non-protective (negative) viral protein sequences. This dataset facilitates the development of machine learning models for predicting viral protective antigens, which are crucial for reverse vaccinology and the development of effective vaccines against viral pathogens.

### Source Data

#### Data Collection and Processing

The dataset was compiled through a meticulous process involving the retrieval of viral PAgs with experimental evidence from the [Protegen](https://violinet.org/protegen/) database, followed by computational steps carried out on viral protein sequences in [UniProt](https://www.uniprot.org/) to select non-protective protein sequences.

## Bias, Risks, and Limitations

Given the imbalanced nature of the dataset, with a greater number of non-protective than protective sequences, there's a risk that machine learning models may become biased towards predicting the majority class. To mitigate this, researchers are encouraged to implement strategies such as balanced sampling or weighted loss functions during model training. Additionally, the dataset's focus on viral proteins from specific databases might limit its coverage of all potential protective antigens across the viral kingdom, which should be considered when generalizing findings.

## Citation

**BibTeX:**

```bibtex
@article{salod2023vpags,
  title={VPAgs-Dataset4ML: A Dataset to Predict Viral Protective Antigens for Machine Learning-Based Reverse Vaccinology},
  author={Salod, Zakia and Mahomed, Ozayr},
  journal={Data},
  volume={8},
  number={41},
  year={2023},
  publisher={MDPI},
  doi={10.3390/data8020041}
}
```

**APA:**

Salod, Z., & Mahomed, O. (2023). VPAgs-Dataset4ML: A Dataset to Predict Viral Protective Antigens for Machine Learning-Based Reverse Vaccinology. Data, 8(41). https://doi.org/10.3390/data8020041

## More Information

This dataset is a crucial step towards leveraging machine learning in the field of vaccinology. By providing a high-quality, curated dataset, VPAgs-Dataset4ML facilitates the development of predictive models that can identify promising vaccine candidates, potentially accelerating vaccine development and deployment in response to emerging viral threats.

## Dataset Card Authors

Zakia Salod, Ozayr Mahomed

## Dataset Card Contact

For any inquiries regarding this dataset, please contact Zakia Salod at [zakia.salod@gmail.com](zakia.salod@gmail.com).