Phager - a rapid phage contig predictor using biological feature based machine learning.
Anastasiya Gæde and Thomas Sicheritz-Pontén
Phager was trained on biological features of phage and non-phage genes. The script phager.py takes a raw or gzipped FASTA file as input, performs gene calling, and generates biological features for each gene. These features are then combined into feature triplets, corresponding to gene triplets along a shifting reading frame. The gene feature triplets are then used by a LightGBM model to determine whether the input genome is a phage.
- Python version 3.8 or later
- Biopython version 1.79 or later
- Pandas
- LighGBM
conda create -n phager pandas
conda activate phager
pip install tqdm icecream colorama pyrodigal biopython more_itertools lightgbm scikit-learn pyarrow
git clone https://github.com/ku-cbd/phager.git
cd phager
python phager.py -h
python phager.py -a example/example_contigs.fasta -o myresults -v
Pyrodigal (Larralde, 2022), a Python library binding to Prodigal (Hyatt et al., 2010). (https://github.com/althonos/pyrodigal)
Sirén,K., Millard,A., Petersen,B., Gilbert,M.T.P., Clokie,M.R.J. and Sicheritz-Pontén,T. (2020) Rapid discovery of novel prophages using biological feature engineering and machine learning. 10.1101/2020.08.09.243022. https://www.biorxiv.org/content/10.1101/2020.08.09.243022v1.abstract
Large-scale analysis of bacterial genomes reveals thousands of lytic phages
Alexander Perfilyev, Anastasiya Gæde, Steve Hooton, Sara A. Zahran, Panos G. Kalatzis, Caroline Sophie Winther-Have, Rodrigo Ibarra Chavez, Rachael C. Wilkinson, Anisha M. Thanki, Zhengjie Liu, Qing Zhang, Qianghua Lv, Yuqing Liu, Adriano M. Gigante, Robert J. Atterbury, Bent Petersen, Andrew D. Millard, Martha R. J. Clokie & Thomas Sicheritz-Pontén
Nature Microbiology volume 11, pages42–52 (2026); doi: https://doi.org/10.1038/s41564-025-02203-4