Research On Construction Of MicroRNA-Disease Bi-layer Network And MicroRNA-Disease Associations Prediction Method

Since the discovery of lots of microRNAs in plants,algae and animals,biomedical researchers have shown great interest in studying them.MicroRNA,as a non-coding RNA with regulatory function,is closely related to cell differentiation,proliferation,aging and apoptosis.Further research on microRNA will affect many basic biomedical fields,such as cell biology and immunology.With in-depth research,microRNA has been proved to be associated with multiple complex human diseases,such as cardiovascular,cerebrovascular diseases,nervous system diseases and various cancers.Recognition of disease-related microRNAs has become an important research topic in the field of biomedicine.The prediction of disease-related microRNA would consume a lot of human and material resources by clinical trials.Therefore,the prediction of disease-related microRNA has become a new hotspot by computational methods.The increasing abundance of biomedical data and the updating of data mining technology provide a new opportunity for the study of predicting disease-related microRNA.From the perspective of bioinformatics,this paper takes microRNA as the research object and predicts the associations between microRNAs and diseases by computational method.The main research work is summarized as follows:(1)There are many existing methods for calculating disease similarity and microRNA similarity,but most models construct similarity network by single similarity.In this paper,we construct a bi-layer microRNA-disease network by different similarity computations.Experiments show that building a bi-layer network based on multiple similarities can improve the performance of the computational model.(2)This paper proposes a method to predict the relationship between microRNAs and diseases based on projective non-negative matrix factorization.The bi-layer relational network is transformed into the original relational matrix,which is then decomposed by using projective non-negative matrix factorization model.Compared with the state-of-the-art methods,the results of 5-fold cross validation showed that our method has higher AUC.Case studies of two complex human diseases also show the superiority of our model.(3)This paper proposes a general framework called NCF(Neural network-based Collaborative Filtering).A model combining the generalized matrix factorization and the multi-layer perceptron is an effective example of the NCF framework,namely the Neural Collaborative Filtering Method(NCFM).NCFM not only realizes matrix factorization,but also uses multi-layer perceptron to enhance nonlinear modeling ability of NCFM model.In addition,this paper uses a more appropriate loss function to optimize the model,this loss makes the ranking of known relationships higher than the unknown relationship.Experiments show that our model is superior to the most advanced algorithms.