Markov random fields (MRF) have been widely used in many different fields including image analysis and theoretical physics. We previously used MRF to predict protein function based on protein interaction networks and individual features with excellent performance. High-throughput metagenomic sequencing technologies have profoundly increased our ability to identify viral genomic sequences without isolation. With the discovery of new viruses, one of the most fundamental challenges is to predict their hosts. We recently developed an integrated MRF model for predicting virus-host interactions based on networks for virus-virus similarity, virus-host similarity, and known virus-host interactions. We used the integrated model to predict hosts of viruses in many different environments yielding important insights into virus-host interactions.