Study On Prediction Of MicroRNA-Disease Association Based On Heterogeneous Data

MicroRNAs(miRNAs)are a member of the large family of non-coding RNAs,and they can affect many biological processes and even affect tissues,organs,systems and even living organisms.Therefore,functional disorders associated with miRNAs are closely related to many human diseases,including cancer.We hope to find and understand more miRNAs associated with human diseases quickly,so as to help treat diseases and even "prevent diseases" through the detection of miRNAs.Because the traditional "wet experiment" is time-consuming and without direction to a certain extent,it is helpful for the future "wet experiment" to use the existing data to predict the possible miRNA-disease associations.On the basis of previous work,a heterogeneous label propagation approach was proposed in this paper to explore the potential associations between miRNA and disease(HLPMDA)by integrating multiple heterogeneous data.First,the data about miRNA,disease and lncRNA was collected.Second,based on the data of miRNA-disease association,long non-coding RNA-disease association,long non-coding RNA-miRNA interaction,miRNA functional similarity and disease semantic similarity,five networks about miRNA-disease,long non-coding RNA-disease,long non-coding RNA-miRNA,miRNA similarity and disease similarity were constructed and integrated into a heterogeneous network.Third,based on this heterogeneous network,the miRNA similarity matrix and disease similarity matrix are updated by projected miRNA and disease topology similarity matrix,respectively.Last,label propagation algorithm can be used on the heterogeneous network and predict potential miRNA-disease associations.As to the experimental results,HLPMDA achieved reliable predictive performance in global and local leave-one-out cross validations(AUC values were 0.9232 and 0.8437,respectively).Moreover,the AUC value and standard deviation were 0.9218±0.0004 in the five-fold cross validation.In addition,case studies of three important human diseases were analysed.Based on previous experiments and literature reports,the top 50 miRNAs of these diseases have been verified one by one.94%,98% and 92% of top 50 predictive results of esophageal neoplasms,breast neoplasms and lymphomas were successfully validated.HLPMDA was also compared with a series of other previous prediction models.On the whole,the predictive performance of HLPMDA surpasses the ten mainstream and classical models of miRNA-disease association prediction,such as PBMDA,MCMDA,Max-Flow,HGIMDA,RLSMDA,HDMP,WBSMDA,MirAI,MIDP and RWRMDA.A website based on HLPMDA for predicting miRNA-disease association has also been developed,which improves the convenience of using the results of this paper to a certain extent.