The human genome project has been completed with a huge success, but theidentification of genes associated with hereditary disorders is a fundamental challengein human health. It is of great importance in improving medical care and a betterunderstanding of gene functions to find disease-causing genes.Genes causing the same or similar diseases tend to lie close to one another in anetwork of protein-protein or functional interactions, which is the modular property ofhuman genetic diseases and have been studied in system biology. Based on this property,many predicting algorithms have been designed and applied into the study of complexdiseases, using the Protein-Protein Interaction networks, disease similarity networks anddisease-gene networks. However, most of these algorithms are higher in timecomplexity.In this work, we present a random walk-based method for predication of diseasegenes by using of a global network distance measure. The method is based on genesfunction consistency and the topological properties of Protein-Protein Interactionnetworks. Furthermore, it is an improved algorithm for computing approximatePageRank vectors which allow us to find such a vector in polynomial time. Bycombining human Protein-Protein Interactions and disease phenotype similarity data,every disease can get a PageRank vector. Each element of the vector represents thescore of the candidate genes. We prioritize all the score to predict candidate genes. Theexperimental results show that the algorithm has a good performance in the algorithmefficiency and the accuracy of the prediction results. |