The Study Of Predicting Disease Genes Based On Interaction Network

Abstract:Complex human diseases are caused by a number of mutated genes which regulate the disease development and affect human physiological activity by interacting with each other. With the rapid development of the high throughput technology and the abundance of biological information, it makes possible to predict these disease related genes based on all kinds of biological information such as interaction network data which is crucial to promote the disease diagnosis and treatment. This paper mainly focuses on predicting disease genes based on interaction network and its main research results list as below:Firstly, a human interaction network (HPIN) is built by integrating the interacted data from multiple databases. Several centrality metrics are adopted to analyze the centrality of genes in human interaction network. By combining the disease gene information from disease database OMIM, the differentiation of centrality between disease genes and others in human interaction network are revealed. The analysis results indicate that average BN value of disease genes is higher than that of others genes, and they also tend to rank higher based on DC and PageRank.Secondly, an improved PageRank method based on weighted interaction network (Weighted PageRank, WPR) is proposed to predict the disease genes in the weighted human disease network. Based on the weighted disease network constructed by the DNA methylation and aberrant methylation, the pagerank values of the genes in the network are calculated. Then the genes with distinguished pagerank values are picked out by compared with genes in1000equal-weighted networks which are built randomly, and the validation are confirmed by querying the literature information of the gene signature library GeneSigDB. The analysis results of lung cancer gene prediction show that the WPR has higher efficiencies and accuracies than Degree Centrality method in weighted network(Weighted Degree Centrality, WDC).Thirdly, for the difference between human tissues, tissue-Specific WPR (SWPR), a disease gene prediction method, is proposed based on several tissue-specific weighted networks. The tissue-specific weighted networks are generated by a Node Removal method based on gene expression data and human interaction network. WPR algorithm is used to recognize the disease genes in tissue-specific weighted networks of different diseases. By comparing the genes prediction of lung cancer and leukemia by WDC, WPR and SWPR, the analysis results reveal that the method of SWPR have more significant advantages in efficiencies and accuracies.To sum up, from the perspective of interaction network, this paper has analyzed the centrality of disease genes and adopted two kinds of method for their prediction. The results will provide positive references for the biologists in biochemical experiments.There are6figures,4tables and136references in this paper.