Research On Prediction Methods Of Disease-related MiRNA Based On Multi-Kernel Fusion

MicroRNAs(miRNAs)are a class of non-gene-encoding small molecule RNAs that play an important role in regulating gene translation and expression at the post-transcriptional level.It effectively induces the degradation of target mRNA(messenger RNA)or inhibits translation and expression of mRNA through biological methods of base complementary pairing.A recent series of studies suggest that miRNAs play an important regulatory role in a variety of biological processes such as the life course of cells and the growth and development of organs.Therefore,the abnormal regulation of miRNA has also led to the occurrence and development of many complex human diseases.In order to effectively overcome the shortcomings and limitations of traditional biological experiments,in recent years,various bioinformatics-based methods have been proposed to predict the association between miRNAs and diseases.At present,the research on the association between miRNA and disease is mainly focused on the research based on network structure and machine learning algorithms.The first method is the self-weighted multi-kernel multi-label learning framework SwMKML.SwMKML adaptively learns the optimal kernel and similarity matrix of two spaces.According to the graph multi-label learning,the prediction scores of two spaces are updated simultaneously.SwMKML uses self-weighted to solve the problem of selecting the optimal kernel in previous experiments.The advantage of the SwMKML is that the association can be predicted using only one association matrix,and new diseases can be predicted.The second method is the nearest neighbor graph regularization matrix factorization model N2 GRMF.N2GRMF is a computing model combining network information and graph regularization,which enhances the generalization ability.This model uses an improved matrix factorization method for manifold learning to improve the accuracy of miRNA-disease association prediction.In addition,N2 GRMF adds a weight matrix to help find potential feature matrix to better predict disease-related candidate miRNAs.In the experimental part,multiple indicators were used to measure the prediction performance of SwMKML and N2 GRMF,and case studies were performed on various human diseases to further judge the performance and feasibility of these models.Experimental results show that both methods are reliable and effective computational models,which can provide some help for further research on the relationship between miRNA and disease.