Clustering PPI Networks of Mixed Host-Pathogen Data Using Biased Repeated Random Walks

Clustering protein-protein interaction (PPI) network data yields groups of proteins that are biochemically involved. Most existing clustering methods treat all the proteins in a PPI network equally. However, analyzing host-pathogen networks requires identification of clusters that represent the interactions between the set of pathogen proteins and the set of host proteins. For studying HIV-human protein-protein interactions, we thus need to identify clusters with the specific composition of at least one virus protein per cluster. Towards this goal, we describe a novel clustering method that focuses on the key virus proteins in a host-pathogen PPI network and utilizes the notion of random walks biased towards specific connectivity configurations. The proposed method finds host-pathogen protein clusters with high accuracy and improves upon the results obtained with other methods at the state-of-the-art.