---
license: cc-by-nc-4.0
language:
- en
tags:
- genomics
- phage-prediction
- bioinformatics
dataset_info:
  features:
  - name: segment_id
    dtype: int64
  - name: contig_id
    dtype: string
  - name: segment_start
    dtype: int64
  - name: segment_end
    dtype: int64
  - name: L
    dtype: int64
  - name: segment
    dtype: string
  - name: label
    dtype: string
  - name: y
    dtype: int64
  splits:
  - name: sample_test_L512
    num_bytes: 5940699
    num_examples: 10000
  - name: sample_test_L1024
    num_bytes: 11060262
    num_examples: 10000
  - name: sample_test_L2048
    num_bytes: 21299753
    num_examples: 10000
  download_size: 18212368
  dataset_size: 38300714
configs:
- config_name: default
  data_files:
  - split: sample_test_L512
    path: data/sample_test_L512-*
  - split: sample_test_L1024
    path: data/sample_test_L1024-*
  - split: sample_test_L2048
    path: data/sample_test_L2048-*
---


# Dataset Card for Phage Prediction Dataset

## Dataset Description

To train and assess our prediction models, we assembled a comprehensive phage sequence database from diverse sources.
As of July 9, 2023, we procured viral sequences and annotations from the RefSeq database. By isolating entries labeled 'phage', we obtained 6,075 contigs. 
Our database was further enriched with the inclusion of the TemPhD database, adding another 192,326 phage contigs extracted from 148,229 assemblies.

To address sequence redundancy present in both the RefSeq and TemPhD databases, we applied the CD-HIT algorithm (using CD-HIT-EST with a default word size of 5). 
While several clustering thresholds (0.99, 0.95, 0.90) were experimented with and found to produce similar outcomes, we settled on a threshold of 0.99. 
This process resulted in a refined set of 40,512 distinct phage sequences, with an average length of approximately 43,356 base pairs, culminating in a total of 3.5 billion base pairs.
Notably, these sequences target a wide spectrum of 660 bacterial genera. Subsequent to sequence curation, phage sequences were mapped to their respective bacterial hosts.

It is a sample dataset consisting of 10,000 segments, representing random samples and segments of phage genomes.
The full dataset is available on [Zenodo](https://zenodo.org/records/10057832).

## Features
- **Phage-Host Associations**: Our dataset represents bacteriophages and their bacterial hosts.

- **Balanced Representation**: The dataset is structured to mitigate bias by evenly representing phages and their hosts across various genera, incorporating reverse-complement sequences for completeness.

- **Dataset Composition**: The final collection includes sequences of varying lengths to accommodate different research needs, with a balanced distribution across training, validation, and testing sets.

- **Sampling Strategy**: To ensure a comprehensive yet manageable dataset, we performed undersampling and upsampling techniques, creating a diverse array of sequence lengths and ensuring no overlap between training and testing sets at the species level.

### Dataset Structure

The dataset is divided into three subsets based on segment lengths: 512, 1024, and 2048 base pairs. These subsets are named `sample_test_L512`, `sample_test_L1024`, and `sample_test_L2048`, respectively.

#### Data Fields

- `segment_id`: Unique identifier for each genomic segment.
- `contig_id`: Identifier for the contig from which the segment is derived.
- `segment_start`: Start position of the segment in the contig.
- `segment_end`: End position of the segment in the contig.
- `L`: Length of the genomic segment (512, 1024, or 2048).
- `segment`: The genomic sequence of the segment.
- `label`: Classification label (e.g., 'phage').
- `y`: Binary label (1 for phage, 0 for non-phage).

### Data Splits

The dataset is structured as follows:

- `sample_test_L512`: Test set with segment length of 512.
- `sample_test_L1024`: Test set with segment length of 1024.
- `sample_test_L2048`: Test set with segment length of 2048.

## Dataset Creation

### Source Data

The dataset is compiled from diverse genomic sources, with a focus on phage sequences and annotations from the RefSeq database and a dataset validated through the TemPhD method. Redundancy in sequences is addressed using the CD-HIT algorithm.


## Contact Information

For any questions, feedback, or contributions regarding the datasets or ProkBERT, please feel free to reach out:

- **Name**: Balázs Ligeti
- **Email**: obalasz@gmail.com

We welcome your input and collaboration to improve our resources and research.



## Citation

```bibtex
@Article{ProkBERT2024,
  author  = {Ligeti, Balázs et al.},
  journal = {Frontiers in Microbiology},
  title   = {{ProkBERT} family: genomic language models},
  year    = {2024},
  volume  = {14},
  URL     = {https://www.frontiersin.org/articles/10.3389/fmicb.2023.1331233},
  DOI     = {10.3389/fmicb.2023.1331233}
}