Abstract: The increasing amount of protein-protein interaction (PPI) data has enabled us to predict protein complexes．Due to the limitation of experimental conditions and techniques,there is a lot of noise in the PPI networks．To reduce the negative effects of noise on protein complex prediction,a new improved method named WPC (Weighted-network based method for Predicting protein Complexes) was proposed．Given a selected node,candidate set consists of all neighbors of the node and neighbor set consists of neighbors of all nodes in the candidate set．If the weighted ratio of a node between the candidate set and the neighbor set is lower than a threshold,the node is removed from the candidate set．After repeating the process for all nodes in the candidate set,the candidate set is represented as a complex．For a node not being included in any complexes,if its average weighted degree within a complex exceeds a self-adjustment threshold,WPC adds the node to the complex．A comprehensive comparison among the competitive algorithms and WPC was made．Experimental results show that WPC outperforms the state-of-the-art methods．

Key words: Average weighted degree,Protein complex,Protein-protein interaction network,Weighted ratio