Research On The Relationship Between Human MiRNA And Disease Based On Heterogeneous Networks

In bioinformatics,the identification of associations between mi RNAs and diseases is a hot direction.Increasing evidence has suggested that miRNAs may function as positive regulators at the post-transcriptional level,which have close associations with human diseases.Hence,the identification of associations between miRNAs and diseases at the systematic level is crucial for understanding the pathogenesis of diseases.However,only a few of them have been studied in detail and have their association with diseases manifested due to the limitation of experimental technique such as time-consumption and high cost.With a large number of miRNAs being identified,various publicly available databases have been constructed to store meaningful information of miRNA molecules.The main work of this paper are summarized as follows:1.To resolve the limitation that individual source of miRNA target data tends to be incomplete and noisy,this thesis proposes a network-based computational method called self-weighting for integrating multiple data source.Afterwards,a bipartite phenotype-miRNA network(BPMN)is constructed,which incorporates known disease-miRNA interactions as well as disease phenotype similarities and miRNA functional similarities.Finally,extend random walk restart algorithm on the bipartite network to predict novel disease-miRNA associations.Compared with other methods,this method can achieve a higher AUC of 0.801 by using the leave-one-out cross-validation experiment according to known disease-related miRNAs from HMDD.Then the method systematically prioritize miRNAs for 11 common diseases and the case study of colon cancer showed that this method can uncover potential disease miRNA candidates.2.Considering that miRNA functional similarity scores and disease phenotype similarity scores do not cover all the miRNAs and disease,and the accurate calculation of miRNA(disease)network topological similarity is fundamental for constructing a network prediction model to predict potential disease-miRNA associations.This paper choices Gaussian Interaction Profile(GIP)kernel,also known as the Radial Basis Function(RBF)kernel for calculating the network topological similarity.Then using a weighted average to calculate the new integrated similarity scores.Finally,extend random walk restart algorithm on the bipartite network to predict novel disease-miRNA associations.Applied this method on 3644 known diseae-miRNA associations based on leave-one-out cross-validation,and achieved an area under the ROC curve of 0.894,which is higher than existing methods.In addition,the case study of breast cancer showed that this method can uncover potential disease miRNA candidates.