Code and Datasets for "PHIAF: Prediction of phage-host interactions with GAN-based data augmentation and sequence-based feature fusion"
Menglu Li (mengluli@foxmail.com) and Wen Zhang (zhangwen@mail.hzau.edu.cn) from College of Informatics, Huazhong Agricultural University.
- data/phage_dna_norm_features is a set of files with features encoded by DNA sequences corresponding to all phages.
- data/host_dna_norm_features is a set of files with features encoded by DNA sequences corresponding to all hosts.
- data/phage_protein_normfeatures is a set of files with features encoded by protein sequences corresponding to all phages.
- data/host_protein_normfeatures is a set of files with features encoded by protein sequences corresponding to all hosts.
- data/data_pos_neg.txt is the dataset used to train and test prediction model, which contain 312 positive and 312 negative samples (304 phages and 235 hosts).
- data/alldata.txt is phage-host interactions extracted from databases, which contain 5399 interactions between 5331 phages and 235 hosts.
The code has been tested running under Python 3.7.9. The required packages are as follows:
- numpy == 1.19.1
- pandas == 1.1.3
- biopython == 1.78
- torch == 1.4.0+cpu
- keras == 2.3.1
- scikit-learn == 0.23.2
- tensorflow == 1.15.0
git clone https://github.com/mengluli-web/PHIAF
cd PHIAF/code
python generate_data.py ####using GAN to generate pseudo positive samples
python main.py
Users can use their own data to train prediction models.
For new host/phage, users can download the DNA and protein sequences from the NCBI database, and use code/compute_dna_features.py and code/compute_protein_features.py to compute the features derived from DNA and protein sequences.
Note:
In code/compute_dna_features.py, users need to install the iLearn tool [https://ilearn.erc.monash.edu/ or https://github.com/Superzchen/iLearn] and prepare .fasta file, this file is DNA sequences of all phages/hosts. (when you use iLearn to compute the DNA features, you should set the parameters k of Kmer and RCKmer as 3, not 2.)
In code/compute_protein_features.py, users need to prepare a .gb file of every phage/host.
Then users use generate_data.py and main.py to predict PHI.
Please feel free to contact us if you need any help.