Heterogeneous Information Network-Based CircRNA Disease Association Prediction

Circular RNA(circRNA)is a class of non-coding RNA which widely exists in various organisms.It has become a hot topic in RNA research area.Ongoing studies continue to demonstrate its critical functions in cell physiology.Currently,there is no systematic study on the prediction of associations between circRNA and disease.In this thesis,we proposed to predict disease-related circRNAs with heterogeneous information network-based methods.Based on the fact that circRNA can act as microRNA(miRNA)sponge,the heterogeneous information network among circRNA,miRNA and disease were constructed.Specifically,first we constructed circRNA co-expression network with the use of next generation sequencing data from ENCODE project;then we proposed a novel method to infer the probability of a circRNA to act as miRNA sponge,and further derived the circRNA-miRNA interaction network;In addition,we proposed to construct miRNA functional similarity network with the direct use of miRNA-target list deposited in miRTarBase,and verified the feasibility of this method;Finally,we constructed the disease network and obtained disease-related miRNAs by cleaning data from related database and documents,respectively.After the construction of heterogeneous information network,we derived different topological features based on the information of various meta-path.Similarity measures of ’path-count’ and ’random-walk’ were used to calculate the relevance between circRNA and disease in this step.Finally,several machine learning algorithms were implemented to predict the potential associations between circRNA and disease based on the obtained features.The performance of these algorithms are compared on different metrics.