Flow Model of the Protein-protein Interaction Network for Finding Credible Interactions

Large-scale protein-protein interactions (PPIs) detected by yeast-two-hybrid (Y2H) systems are known to contain many false positives. The separation of credible interactions from background noise is still an unavoidable task. In the present study, we propose the relative reliability score for PPI as an intrinsic characteristic of global topology in the PPI networks. Our score is calculated as the dominant eigenvector of an adjacency matrix and represents the steady state of the network flow. By using this reliability score as a cut-off threshold from noisy Y2H PPI data, the credible interactions were extracted with better or comparable performance of previously proposed methods which were also based on the network topology. The result suggests that the application of the network-flow model to PPI data is useful for extracting credible interactions from noisy experimental data.