Protein function prediction based on k-cores of interaction networks

This paper presents a method for prediction of protein functions based on protein-protein interaction (PPI) networks. It can be hypothesized that unknown function proteins that form a densely connected subgraph with proteins of a particular function are likely to belong to the same functional group. The proposed approach utilizes this concept by determining k-cores of strategically constructed subnetworks. For the purpose of prediction, not only the interactions of function-unknown proteins with function-known proteins but also the interactions of function-unknown proteins with function-unknown proteins were considered. The proposed method has been applied to the PPI network of Arabidopsis thaliana consisting of 3,118 interactions and 1,302 proteins and functions of 88 unknown proteins have been predicted. The proposed method is a general one and can be applied to other organisms too. It was found that in the context of present data k-cores with the value of k equal to 3 or more produce statistically significant predictions.