Predicting disease-related genes by path structure and community structure in protein-protein networks

Abstract

Network-based computational approaches in the prediction of genes that are associated with diseases are of considerable importance in uncovering the molecular basis of human diseases. Here, we proposed a novel disease-gene-prediction method by combining path-based structure with community structure characteristics in human protein-protein networks. A new similarity measure was first proposed that is based on the path and community structures of networks and leverages community structures for disease-gene prediction. Then, the distinguishing capacity of the methods to identify disease genes from non-disease genes was assessed statistically to analyze their ability to predict disease genes. Finally, the new method was applied to disease-gene prediction for several datasets, and the performances of the measures in disease-gene prediction were analyzed, with an emphasis on assessing the effect of community structure on the predictive performance. The results indicated an ability of the new method to predict disease-genes in several networks and within several disease classes. Further, the results reported here confirm that the incorporation of community structures can indeed improve the performance of disease-gene prediction methods.