Disease Gene Prioritization Based On A Tissue-specific Network Model

Prediction and identify of disease genes could help to understand the mechanism of diseases,and is important for diagnosis and treatment of diseases.Traditional biological experiment method to identify disease gene is expensive,time consuming and has technical difficulties.Therefore,predicting the disease genes with bioinformatics method has become an important method to predict disease genes.So-called disease gene prioritization,refers to the process of arranging possible disease causing genes in order of their likelihood in disease involvement,in order to measure the possible pathogenic possibility or possibility participate in a particular disease process,thus biological experiments can be done for validation.At present,people have proposed various types of disease gene prioritization method.One of the mose important category is based on the network model,and the experimental results show that this kind of method has high accuracy for disease gene prioritization.The main idea of network-based disease gene prioritization is as described below,first of all,we build an appropriate network in some way,then the information of known disease genes are mapped in the network,according to a certain distance or similarity measure between candidate genes and the known disease genes in the network,all the candidate genes can be scored and ranked.Most of the current network-based methods based on the following assumption,namely all diseases are associated with the same molecular interaction network.However,many of the diseases tend to be appear in a particular tissue,and different tissues relate to different molecular interaction networks.This paper intends to use a new network model for disease gene prioritization,the network model based on this idea:for different diseases,do not use the same molecular interaction network,but for each disease,determine the tissue which most likely to appear,then the disease is associated with the most relevant tissue-specific molecular interaction network,and the subsequent steps such as random walk will be done.In this paper,the random walk method uses a new metric to measure the distance between the candidate genes and the known disease genes.Then random walk is done on the network and we obtain the rankings of disease genes as a result,first of all,the efficiency of the method and parameters are analyzed,find that the random walk method is of high efficiency,and can be convergence within 10 iteratives.Next the method was compared with the PRINCE method and BlockRank method,demonstrate that our method is more accurate than other methods in gene priorization using cross validation;In the next step,random walk is executed in the global protein-protein interaction network,based on the results of cross validation,we can come to a conclusion that tissue-specific protein-protein interaction network can improve the accuracy of the disease gene ranking effectively,for amyotrophic lateral sclerosis corresponding AUC value can be improve from 0.758 to 0.876,for thyroid hormone resistance syndrome corresponding AUC value can be improve from 0.788 to 0.852;Finally,the experimental results are verified through the literature use pancreatic cancer as an example,and we find that in the top 10 genes,six genes associated with pancreatic cancer have been confirmed in literature,so our our method is reliable for disease gene priorization.