Research On MiRNA-disease Association Prediction Algorithm

Biological research and scientific experiments show that miRNA affects the occurrence and development of complex human diseases.Accurate identification of potential miRNA-disease associations can not only enhance the understanding of the molecular mechanisms and pathogenesis of diseases,but also promote the diagnosis and prevention of human diseases.Since it is time-consuming and laborious to identify the miRNA-disease association with traditional biological experimental methods,computational methods for predicting the potential miRNA-disease association have attracted much attention.In recent years,many association prediction algorithm models based on similarity and machine learning have been proposed one after another,but the quality of the constructed network and the characterization ability of the extracted features need to be improved.In response to this problem,this paper applies different network embedding algorithms to the establishment of miRNA-disease association prediction algorithm model,and points out that the network embedding algorithm based on meta-path can better capture the latent structural and semantic features of the network.In order to further capture the latent semantic features in the network,a new association prediction algorithm model was designed by combining the attention mechanism in natural language processing with the core idea of random walk in network embedding algorithm from the perspective of algorithm design.The main research contents and achievements are as follows:Aiming at the problem that the network quality is not high enough,a variety of biological data such as miRNA-disease association data,miRNA functional similarity data,disease semantic similarity data,similarity data based on Gauss interaction profile kernel were integrated to construct a high-quality miRNA-disease association heterogeneous network.In view of the problem of insufficient feature representation capabilities,the relevant network embedding algorithm is used to capture the global structural similarity information and semantic similarity information between nodes in the network on the heterogeneous network to generate node embedding vector with strong representation capabilities,combined with random forest algorithm to achieve the construction of miRNA-disease association prediction algorithm model.Compared with other association prediction models based on network embedding algorithms,the association prediction model constructed using the meta-path based network embedding algorithm achieves the highest AUC of 0.919,which indicates the advantages of network embedding algorithm based on meta-path in mining potential structural and semantic features of the network.A random walk algorithm based on meta-path is used to generate node sequences,and the attention mechanism in natural language processing is introduced to capture the semantic similarity information and global structure similarity information between the nodes in the network,the miRNA-disease association prediction algorithm model ADMDA based on attention mechanism is proposed.In the process of model validation and evaluation,ADMDA achieved the highest AUC of 0.945 compared with several other classical algorithm models under the framework of five cross-validation,and achieved excellent performance results in the case study of kidney neoplasms and breast neoplasms.