PHIHNE: predicting Phage-Host Interaction through Heterogeneous Network Embedding

Abstract

The volumes of novel phages obtained by metagenomics demand computational tools to predict phage–host interactions. Compared with the experimental approach, the identification of phage–host interactions by computational method can save time and reduce costs. In this paper, we present a computational method for predicting potential phage-host interactions by network fusion and graph mining, named PHIHNE. Unlike existing methods, PHIHNE constructs two different viral host heterogeneous networks by similarity network fusion and graph embedding techniques. Then, PHIHNE introduces two meta-path scores to extract features from each viral host heterogeneous graph. Based on this graph mining approach, a hybrid feature vector of phage-host pairs can be obtained to predict potential phage-host interactions using a machine learning classifier. PHIHNE is validated on four datasets and its performance shows the potential of PHIHNE in predicting phage-host interaction. Some of the novel phage-host interactions predicted by PHIHNE have been verified by existing in biological experiments.