Detecting the hosts of bacteriophages using GCN-based semi-supervised learning

Motivation: Bacteriophages (aka phages) are viruses that infect bacteria and archaea. Thus, they play important regulatory roles in natural and host-associated ecosystems. As the most abundant and diverse biological entities in the biosphere, phages have received increased attention in their research and applications. In particular, identifying their hosts provides key knowledge for their usages as antibiotics. High-throughput sequencing and its application to the microbiome have offered new opportunities for phage host detection. However, there are two main challenges for computational host prediction. First, the known phage-host relationships are very limited compared to sequenced phages. Second, although the sequence similarity between phages and bacteria has been used as a major feature for host prediction, the alignment is either missing or ambiguous for accurate host prediction. Thus, there is still a need to improve the accuracy of host prediction. Results: In this work, we present a semi-supervised learning model, named HostG, to conduct host prediction for novel phages. We construct a knowledge graph by utilizing both phage-phage protein similarity and phage-host DNA sequence similarity. Then graph convolutional network (GCN) is adopted to exploit phages with or without known hosts in training to enhance the learning ability. During the GCN training, we minimize the expected calibrated error (ECE) to ensure the confidence of the predictions. We tested HostG on both simulated and real sequencing data and the results demonstrated that it competes favorably against the state-of-the-art pipelines.