Research On MiRNA-disease Association Model Based On Autoencoder

MiRNAs,as small molecules with regulatory functions,participate in several biological processes such as tumor development,cell growth,immune response and metabolism,and their abnormal expression often accompanies the emergence of specific diseases.Therefore,the identification of disease-related miRNAs is conducive to deepen the understanding of disease pathogenesis and promote the early detection of cancer and disease-targeted therapy.In the method of discovering new miRNA-disease associations,traditional biological experiments can obtain accurate results,but there are disadvantages such as special equipment requirements,complicated operations and high cost.With the generation of massive omics data,rapid prediction of potential miRNA-disease associations through data and computational methods has become a new option.Such methods can obtain more accurate experimental targets and obtain associations more efficiently,which makes revealing miRNA-disease relationships by developing systematic and reliable computational models a promising research area.Although a large number of computational models for revealing miRNA-disease relationships have been proposed,the prediction accuracy of existing methods is still low,so it is still very challenging to accurately and efficiently identify miRNA-disease associations.This paper is devoted to rationally utilizing a large number of known heterogeneous biological datasets to develop computational prediction models that can effectively identify miRNA-disease associations to overcome existing problems and provide reliable and specific experimental objects for subsequent biological experiments.The main work is as follows:(1)Aiming at the problem that supervised machine learning prediction models require negative samples without real negative samples,a method for miRNA-disease association prediction(AELPMDA)that fuses graph autoencoder and label propagation is proposed.First,the complex associations between miRNAs and diseases are integrated based on known association information,similarities between them.Second,we use a GCN-based graph autoencoder to learn multiple representations within graphs from miRNAs and diseases to fully extract comprehensive information of nodes.The LP network is then constructed to model the local dependencies of object labels.Finally,the EM algorithm is used for alternate training to obtain the final result.The method is verified by five-fold(ten-fold)cross-validation and case studies on the HMDD dataset,and the experimental results verify the superiority of the AELPMDA method.(2)In view of the problem that insufficient accuracy caused by the high dimension of the original data,a new computational method via deep forest ensemble learning based on auto-encoder(DFELMDA)is proposed to predict miRNA-disease associations.Specifically,a new feature representation strategy is proposed to obtain different types of feature representations(from miRNA and disease)for each miRNA-disease association.Then two types of low-dimensional feature representations are extracted by two deep auto-encoders for predicting miRNA-disease associations.Finally,two prediction scores of the miRNA-disease associations are obtained by the deep random forest and combined to determine the final results.The method is validated on the HMDD dataset using cross-validation and case studies.The experimental results show that the DFELMDA method can effectively predict the potential association of miRNAs with diseases,improve the accuracy of prediction and can be used as an effective computational tool for predicting miRNA-disease associations.