Research On Complex Disease-related MiRNAs Prediction Algorithm And Its Application

The identification of miRNAs related to complex diseases has become an important research topic in the field of biomedicine,which has attracted the interest of researchers in recent years.As a class of non coding RNA with post transcriptional regulation function,miRNA is combined with the 3-UTR of target gene in base pairing to inhibit or degrade target gene expression.There is a lot of evidence that miRNA has been involved in many important biological processes,and its abnormal expression can cause many complex human diseases,including cancer.To explore the potential association between miRNA and disease is helpful to understand the pathogenesis of disease from miRNA level,and provide effective assistance for early diagnosis,treatment prognosis and drug design.However,it takes a lot of time and cost to identify disease-related miRNAs by using biological experiments.It is an urgent need to develop a reliable method for predicting miRNA-disease.The purpose of this paper is to build accurate and effective miRNA-disease association prediction model based on the existing massive biological data and the current popular computing methods,so as to provide guidance for subsequent biological experiments on verifying complex disease-related miRNAs.According to different tasks,different prediction algorithms have been proposed in the paper to identify pathogenic miRNAs and discover association types between disease and miRNAs.We applied the algotithms in complex diseases such as lung cancer,breast cancer and colorectal cancer,to predict disease-related miRNAs and discover new clue for disease treatment.The main work of this paper is summarized as follows:(1)The research background and purposes of this topic have been introduced firstly.We also analyzed overseas and domestic research status on miRNA and cancers,and summarized main problem existed as well as data resources closely related to our research.Then,we analyzed the key problems of similarity network construction in miRNA-disease association prediction research.(2)MiRNA-disease association prediction based on improved low rank matrix recovery method.The accuracy of the existing disease-related miRNAs prediction model is still not high.Many machine learning prediction methods based on supervised learning require negative samples,while limited biological experiments cannot determine the real absence of association between miRNA and diseases,resulting in the problem of negative samples being absent or difficult to obtain.In this paper,a method based on improved low rank matrix recovery(ILRMR)is proposed to predict the association between miRNA and disease.ILRMR algorithm integrates miRNA function similarity,topology similarity and miRNA family information to reconstruct miRNA similarity network,and reconstructs the disease similarity network by integrating the semantic similarity and topological similarity of the disease,then it integrates the similarity information and builds the miRNA-disease association prediction model based on RPCA.The ILRMR algorithm is a global prediction method that can simultaneously predict the potential pathogenic miRNA associated with all diseases.ILRMR does not need negative samples,especially when the sample is sparse,it can still maintain reliable prediction performance.The results of cross validation and case studies show that the predictive performance of the ILRMR algorithm is better than that of the existing methods.(3)The study of miRNA-disease association prediction based on the regularized framework fusion heterogeneous omic data.Some existing prediction models rely too much on the known miRNA-disease association in the process of calculating miRNA similarity and disease similarity,resulting in overestimation of predictive perfomance in LOOCV,and the additional overhead incurred by similarity recalculation.Furthermore,many prediction models can not predict the pathogenic miRNAs associated with isolated disease.Therefore,in this paper,a regularized framework based on information fusion strategy(RLSSLP)is proposed for predicting pathogenic miRNA based on heterogeneous omic data.RLSSLP model measures the similarity between the two miRNAs according to the similarity between the two target gene sets of miRNA,and it does not depend on the known association between miRNA and disease.RLSSLP model takes into account the relationships between disease-gene,gene-gene and miRNA-gene,integration two sub models including regular least square and edge propagation algorithm to construct an efficient pathogenic miRNA prediction model.Meanwhile,it improves the computational efficiency using feature conversion technology.The RLSSLP model is a global prediction method,which can effectively predict isolated disease-related miRNAs as well as new miRNA-related diseases.The results of 10-fold cross validation and case studies of lung cancer,hepatocellular carcinoma,and breast cancer showed that the RLSSLP method showed a reliable predictive performance.(4)MiRNA-disease association type prediction based on hybrid constrained Boltzmann model.At present,most of the methods can only predict the two dollar relationship between miRNA and disease.The rich information about the different association types between miRNA and disease has not been used well in the miRNA-disease prediction.The association of miRNA and disease caused by different types of pathogenic mechanisms can not be well predicted in the existing methods.Based on these considerations,a new model of hybrid constrained Boltzmann model for miRNA and disease association type prediction(HRBM-MD)is proposed in this paper.This model extends the traditional constrained Boltzmann model in some respects,and it constructs a hybrid model of hidden layer from two angles of miRNA and disease,which improves the accuracy of prediction effectively.The HRBM-MD method can effectively predict the type of miRNA-disease association caused by four aspects,such as genetics,epigenetics,circulating miRNA and miRNA-target interactions.It is helpful to understand the pathogenesis mechanism of the disease caused by miRNA.Cross validation and case studies show that HRBM-MD has a reliable predictive performance.