Interaction-based feature selection for predicting cancer-related proteins in protein-protein interaction networks

The task of predicting in a protein-protein-interaction (PPI) network which proteins are involved in certain diseases, such as cancer, has received a significant amount of attention in the literature [1, 4]. Multiple approaches haven been proposed, some based on graph algorithms, some on standard machine learning approaches. Machine learning approaches typically use a feature- based representation of proteins as input, and their success depends strongly on the relevance of the selected features. In earlier work it has been shown that the Gene Ontology (GO) annotations, here called "functions", of a protein have high relevance; whether a protein is likely disease-related depends on its functions. In this paper, we present a new type of GO-based features. These are based not on the function of a single protein, but on pairs of functions that occur on both sides of an edge in the PPI network. We call them interaction-based features. We show that these features provide better predictive accuracy than the functions of the target protein itself.status: publishe