Association Prediction Of Circrnas And Mirnas Based On Network Embedding

Circular RNA(circRNA)is a non-coding RNA with a closed loop structure.A large number of biological experiments have shown that circRNA can be used as a micro RNA(miRNA)sponge,which can indirectly regulate the expression of miRNA target genes and thus play an important role in the formation,detection,diagnosis,or treatment of some diseases.Therefore,the correct identification of the associations between circRNA and miRNA can not only help understand the complex disease mechanism,but also help the diagnosis and treatment of the disease.In order to better preserve the circRNA-miRNA-cancer association,this paper constructs a comprehensive database(circR2Cancer)that stores the circRNA-miRNA-cancer association.Compared with the existing circRNA and disease association databases,circR2 Cancer not only stores more association data and rich basic information,but also provides more convenient data retrieval methods for users.In addition,on the basis of the existing data,this paper proposes two prediction models of circRNA and miRNA potential associations based on computational methods.The first model is NECMA,which uses the network embedding algorithm for the feature extraction of circRNA and miRNA for the first time,and then implements the potential association prediction between circRNA and miRNA through neighborhood regularization logic matrix decomposition and inner product.The 10-fold cross-validation results show that compared to the other three advanced association prediction models(RWRLnc D,NCPLDA,and LRLSLDA),the NECMA model achieves higher AUROC and AUPR values.Moreover,case studies show that the NECMA model can correctly identify the potential association between circRNA and miRNA.The second is the prediction model of potential association between circRNA and miRNA based on multi-biological association data(IIMCCMA).Based on the fusion of multiple association data related to circRNA and miRNA,this model uses an improved inductive matrix completion algorithm to predict the potential association of circRNA and miRNA.In this model,this paper focuses on the influence of fusion features on the model's predictive ability.In addition,the sparseness of the known circRNA and miRNA potential association will negatively affect the subsequent prediction of the potential association between circRNA and miRNA through the inductive matrix completion algorithm.Hence,this paper uses multiple spatial mappings of different dimensions in the matrix completion process to obtain more structural information and uses lowdimensional space vectors of different dimensions to calculate the corresponding circRNA and miRNA association prediction scores.Finally,the prediction scores calculated by the low-dimensional space vectors of different dimensions are integrated to obtain the final potential association score of circRNA and miRNA.In order to verify the influence of the fusion feature and the improved inductive matrix completion algorithm on the model performance,this paper constructs two benchmark models for comparison with IIMCCMA.The 10-fold cross-validation results show that the IIMCCMA model performs better than the two benchmark models.In addition,in order to prove the superiority of the performance of IIMCCMA,this paper compares it with LLCDC,CD-LNLP,RWR and KATZCPDA.The results of the ten-fold cross-validation experiment show that the IIMCCMA model has achieved higher AUROC and AUPR values.As a result,it can be concluded that the IIMCCMA model performs better in the prediction of the potential association between circRNA and miRNA.In addition,the case study results show that the IIMCCMA model can correctly identify the potential association between circRNA and miRNA.