Global Voting Model for Protein Function Prediction from Protein-Protein Interaction Networks

Abstract

It is known that the observed PPI network is incomplete with low coverage and high rate of false positives and false negatives. Computational approach is likely to be overwhelmed by the high level of noises and incompleteness if relying on local topological information.We propose a global voting (GV) model to predict protein function by exploiting the entire topology of the network. GV consistently assigns function to unannotated proteins through a global voting procedure in which all of the annotated proteins participate. It assigns a list of function candidates to a target protein with each attached a probability score. The probability indicates the confidence level of the potential function assignment. We apply GV model to a yeast PPI network and test the robustness of the model against noise by random insertion and deletion of true PPIs. The results demonstrate that GV model can robustly infer the function of the proteins.