A novel predictor for disease-genes based on combination use of topological features in human protein-protein interaction network

Abstract

Generally, complex human diseases are caused by multiple gene mutations, and these mutated genes affect the development of disease through interactions. It is believed that the structure of a network indicates its function, and the stability of the network is determined by some crucial points. In this paper, the centricity of the proteins (genes) in human protein-protein interaction network is considered as an indicator to distinguish disease and non-disease proteins. These features include K, BC, and Disease_C. And we find that disease-genes are more topological important than non-disease genes. Further, SVM classifier base on our topological features is created and gains a good prediction accuracy of 69.33% in 10-fold cross-validation test. The experimental results reveal that these three features can improve the prediction of disease-genes, and it demonstrates that the centricity of the proteins (genes) is helpful for finding key proteins in the PPI network.