Motivation: Protein-protein interaction (PPI) networks are a valuable resource for the interpretation of genomics data. However, such networks have interaction enrichment biases for proteins that are often studied. These biases skew quantitative results from comparing PPI networks with genomics data. Here, we introduce an approach named physical interaction enrichment (PIE) to eliminate these biases.
Methodology: PIE employs a normalization that ensures equal node degree (edge) distribution of a test set and of the random networks it is compared with. It quantifies whether a set of proteins have more interactions between themselves than proteins in random networks, and can therewith be regarded as physically cohesive.
Results: Among other datasets, we applied PIE to genetic morbid disease (GMD) genes and to genes whose expression is induced upon infection with human-metapneumovirus (HMPV). Both sets contain proteins that are often studied and that have relatively many interactions in the PPI network. Although interactions between proteins of both sets are found to be overrepresented in PPI networks, the GMD proteins are not more likely to interact with each other than random proteins when this overrepresentation is taken into account. In contrast the HMPV-induced genes, representing a biologically more coherent set, encode proteins that do tend to interact with each other and can be used to predict new HMPV-induced genes. By handling biases in PPI networks, PIE can be a valuable tool to quantify the degree to which a set of genes are involved in the same biological process.