Construction and application of weighted protein protein interaction network based on multiple views

Predicting protein complexes from protein-protein interaction (PPI) network plays an essential role in exploring the cell organization and detecting protein function or structure. Whereas, the data noise from the PPI network will have an inevitable impact on the prediction result. In order to filter out the data noise, this paper proposed a method which introduced the topological information and biologicalmetrics to build the feature vector for protein couples or the interactions, instead of a single protein, then the cosine theory is applied for measuring the similarity between the feature vector and the unit vector, at last, the similarity value will be considered as the weight of PPI network. In the experiments part, we used five classical prediction algorithms, including IPCA, MCODE, Graph-entropy, COACH and Clique, to detect protein complexes from four types of weighted PPI networks respectively. The results illustrated that our method performs better on filtering out data noise and predicting protein complexes.