MiRNA And Disease Association Prediction Based On Hypergraph

MicroRNAs(miRNAs)are a class of small molecule non-coding RNA with regulatory functions.A large amount of research evidence shows that mutation and disorder of miRNAs are important causes of disease,so identifying disease-related miRNAs has become an important topic in biological research in recent years.However,the high cost,long validation cycle and blindness of traditional biological experimental methods limit the rapid development of miRNA-disease association research.With the gradual accumulation of miRNA-disease association data,researchers have established some highly reliable public databases.These databases provide experimental validated miRNA-disease associations and related biological information.Researchers make full use of the existing database data,design efficient and accurate calculation methods to predict the potential miRNA-disease association,make up for the shortcomings of biological experiment methods,reduce the research cost and shorten the research cycle,provide new ideas for disease research and new theoretical basis for its diagnosis and treatment.At present,most of the computational methods have the disadvantages of low prediction accuracy,inability to predict new diseases,low quality of feature set,and difficult to mine the deep features of nonlinear high-order association.In order to solve these problems,three computational models of miRNA-disease association prediction are proposed by using a variety of biological data to construct a high-quality similarity network.The main works are as follows:(1)HyperGraph for Predicting MiRNA-disease Association(HGMDA).HGMDA model used the semantic similarity of disease,the functional similarity of miRNA and the known association data between miRNA and disease as the input data,and represented the feature vector of miRNA-disease association based on statistical theory,graph theory and matrix factorization.Considering that the hypergraph structure has stronger ability to describe and mine the non-linear high-order association between data samples,this paper used the k-means algorithm to construct a miRNA-disease associations hypergraph.Based on the inductive hypergraph learning,we got the mapping matrix from the feature to the association score,and used this matrix to calculate the association score of an unknown miRNA-disease pair.(2)MiRNA-disease association prediction via hypergraph learning based on high-dimensionality features(HFHLMDA).HFHLMDA took into account the sparsity of similarity information,complements the data with Gaussian kernel interaction profile,and then used the similarity information as the feature vector to learn the mapping matrix based on the improved hypergraph learning model.(3)Multi-Similarity based Combinative Hypergraph Learning for Predicting MiRNA-disease Association(MSCHLMDA).Some diseases or miRNAs still have no similarity information after reconstruction,MSCHLMDA used the nearest neighbor association data to evaluate the unknown association to increase the Gaussian kernel similarity data,then integrated the similarity data of multiple miRNAs or diseases,and designed a combined hypergraph learning algorithm based on the simplified and effective feature composition.By learning a more comprehensive combination mapping matrix,the prediction result was more accurate.All the three methods were evaluated by the leave-one-out cross validation and five-fold cross validation,and further confirmed in the subsequent case studies.Experimental results show that three models such as HGMDA can be effectively used for predicting miRNA-disease associations.