The Research On Algorithm Of Predicting MiRNA-disease Associations By Integrating Multiple Types Of Genomic Data

MicroRNAs(miRNAs)are a class of endogenous single-stranded,non-coding RNAs of approximately 20-24 nucleotides in length,which utilize base pairs to bind to the 3' non-coding region of the target mRNA,thereby degrading or inhibiting translation,and the relevant expression of the target gene is ultimately regulated at the post-transcriptional level.Previous researches have shown that miRNAs played an important role in human life activities.The abnormal expression or mutation of miRNAs could lead to the occurrence of various diseases.Studies have shown that miRNAs have important links with human life activities,and abnormal expression or mutation of miRNAs can lead to various complex diseases.Traditional research mainly uses biological experiments to determine the association between individual entity attributes.Although the accuracy of biological experiments is high,this method has higher requirements for equipment cost and experimental period.Therefore,predicting the association between miRNA diseases by bioinformatics has become a research hot topic in the biomedical field.In this paper,we take the biological networks as the main research object and combine with massive biological data resources and computer related technology to study the association algorithm between miRNA complex diseases.The main research is as follows:Firstly,the existing miRNA disease association recognition algorithms based on biological networks mostly ignore the influence of network topology map feature information.Through analysis of heterogeneous networks consisting of miRNAs and diseases,we found that most experimentally validated disease-miRNA interactions are associated with subgraphs composed of specific neighbor nodes.Based on the hypothesis that most potential disease-miRNA associations are affected by their neighbor subgraphs,in this paper we propose a new miRNA-disease association prediction algorithm based on subgraph extraction.The algorithm first uses the semantic relationship data between diseases and the experimentally verified miRNAdisease association data to calculate similarity for disease as well as miRNA pair and construct a two-layer heterogeneous network.Then,based on the correlation between the nodes in the heterogeneous network,we extract the specific graph information as well as tags of the known association and then train the logistic regression model.Finally,we use this model to predict the association of potential miRNA diseases.The experimental results show that the algorithm has higher prediction performance than the comparison algorithm.Secondly,bio-omics data is accumulating with the development of highthroughput technology.More and more data can mine more potentially valuable information.Traditional miRNA-disease prediction algorithms ignore the existence of potential negative sample relationship,that is,low correlation samples.Therefore,in order to improve the accuracy and efficiency of existing prediction algorithms,we propose a novel miRNA-disease relationship prediction framework based on semisupervised learning model.The framework first uses K-Mediods clustering algorithm and similarity data to cluster miRNA and disease respectively,and selects low correlation samples as negative samples in unknown association samples according to known miRNA-disease associations.Subsequently,we use existing algorithm to predict of the relationship of disease and miRNA.Finally,we combine the predicted score with negative sample data to select and rank candidate miRNAs.The experimental results show that the framework has a significant improvement effect on the prediction performance of existing algorithms.