Disease-gene Prediction Based On Topological Similarity In Human Protein-protein Interaction Network

Disease-gene prediction provides a significant instruction for discovering disease genes in experiment, and is a key step for the rapid identification of disease gene.Recently, the improving protein-protein interaction network provides us with new avenue for elucidating the disease genes directly from the network. However, the current network-based algorithms focus on either the directed interactions among proteins or their topological features, while neglect the important impacts of physical and/or functional module existing in the protein-protein interaction network, so that their success is limited. In this thesis, we start from the community structure corresponding functional modules, and analyzed the statistical properties and community structure of human protein-protein interaction network. It is discovered and proved that the disease genes with the same or similar disease phenotypes have functional relevance unsually form the specific physical or functional modules in the protein-protein interaction network, and locate in the same community. Based on the analysis and consideration, we have put forward a kind of disease-gene prediction method by combining the network topological similarity and community structure, and obtained a good result for prediction. The main results of this thesis are summarized as follows:(1) Based on the human protein-protein interaction network, we adopt the local topological similarity in network to predict the hepatocellular-carcinoma related genes.The cross validation showed that the AUC of every similarity index can exceed 0.7, and22%-29% known disease genes are at top 5%. Due to both low computing complexity and relatively high prediction accuracy, they might be applied to speed up the identification of disease-related genes in experiment.(2) By analyzing the features of community structure of human protein-protein interaction network, we firstly propose a community-based similarity index, and have developed a novel method for disease-gene prediction by reasonably combining it with the path-based similarity. And then, we carried out a statistical analysis of disease genes and non-disease genes in network, and confirmed the feasibility of the community-based disease gene prediction. Finally, we applied this method to predict hepatocellular-carcinoma related genes, and investigated the performance, especially for the influence of community structure on the prediction performance. The results indicate that genes associated with the same or similar diseases commonly reside in the samecommunity of the protein-protein interaction network, and the community structure can greatly enhance the accuracy of the disease-gene prediction.