Predicting Disease Genes from Clinical Single Sample-Based PPI Networks

Experimentally identifying disease genes is time-consuming and expensive, and thus it is appealing to develop computational methods for predicting disease genes. Many existing methods predict new disease genes from protein-protein interaction (PPI) networks. However, PPIs are changing during cells’ lifetime and thus only using the static PPI networks may degrade the performance of algorithms. In this study, we propose an algorithm for predicting disease genes based on centrality features extracted from clinical single sample-based PPI networks (dgCSN). Our dgCSN first constructs a single sample-based network from a universal static PPI network and the clinical gene expression of each case sample, and fuses them into a network according to the frequency of each edge appearing in all single sample-based networks. Then, centrality-based features are extracted from the fused network to capture the property of each gene. Finally, regression analysis is performed to predict the probability of each gene being disease-associated. The experiments show that our dgCSN achieves the AUC values of 0.893 and 0.807 on Breast Cancer and Alzheimer’s disease, respectively, which are better than two competing methods. Further analysis on the top 10 prioritized genes also demonstrate that dgCSN is effective for predicting new disease genes.