A Research On Prediction Method Of Disease Related-miRNA Based On Biased Heat Conduction Recommended Algorithm

Micro RNAs(mi RNAs)are a class of endogenous single-stranded non-coding small RNAs with a length of about 22nt,which usually inhibit gene expression and protein production of the 3~′untranslated regions(UTRs)of their target messenger RNA after transcription.With the rapid development of genome sequencing technology and computer technology,many studies have shown that mi RNAs are involved in a variety of biological processes,such as cell proliferation,cell development,cell apoptosis,cell differentiation etc,and have been proved to be closely related to a variety of human diseases.Studying the association between mi RNA and disease is helpful to comprehend the pathogenesis of disease at the mi RNA level,which can provide effective help for the treatment and prevention of related diseases and the development of related drugs.Therefore,it is necessary to construct an efficient and accurate predictive model to explore the potential association between mi RNA and disease.Relevant scholars have put forward many prediction models,but there are problems such as long time consuming and low prediction accuracy.Considering the limitations of these models,a new clustering algorithm is first proposed in this paper to process the known data sets of mi RNA-disease association and construct binomial graph networks.On this basis,relevant mathematical models were constructed by using the partial heat transfer algorithm of personalized recommendation system to discover the potential association between disease and mi RNA.The specific research contents are as follows:(1)Firstly,mi RNA-disease association data were obtained and pretreated.Secondly,mi RNA-mi RNA similarity and disease-disease similarity were calculated.Then the functional similarity between mi RNAs and mi RNAs,gaussian interaction profile kernel similarity for mi RNAs,semantic similarity between diseases and diseases,gaussian interaction profile kernel similarity for diseases were respectively integrated.(2)Based on the integrated data,a threshold-based clustering algorithm was proposed to redefine the mi RNA-disease association which could be expressed by adjacency matrix.Then these adjacency matrices were used to construct mi RNA-mi RNA bipartite graph network and disease-mi RNA bipartite graph network respectively.(3)Based on the constructed mi RNA-disease bipartite graph network and disease-mi RNA bipartite graph network,and based on the partial heat conduction algorithm in the personalized recommendation system,a new mathematical prediction model was proposed to explore the potential association between mi RNA-disease.The model can be applied to the environment with sparse data,and it has certain reference significance to solve the cold start problem,and the prediction result is more accurate.The AUC values are 0.8890,0.9060 and 0.8931 respectively under the framework of Leave-One-Out Cross Validation,two-fold and five-fold cross validation.In the case analysis of Esophageal Neoplasms,Colonic Neoplasms and Lymphoma,the prediction accuracy was 88%,92%and 92%,respectively.Thence,this model would be a useful calculative resource for potential mi RNA-disease association prediction.