Association Prediction Between LncRNA And Disease Based On Deep Learning

In recent years,with the development of high-throughput technology in the genome transcriptome,a great number of non-coding RNAs have been identified as disease-related.Studies have shown that a variety of biological macromolecules such as long non-coding RNA(lncRNA)are not only widely involved in various basic physiological activities,but also play an important role in the occurrence and development of various diseases.The mechanisms by which lncRNAs regulate physiology and pathology are complex,while traditional biological wet experiments have disadvantages such as high cost and long time.Therefore,predicting the association between lncRNAs and diseases by means of bioinformatics is of great importance to efficiently and economically understand the pathogenesis of diseases at the molecular level.At present,many computational methods for lncRNA disease association prediction have been proposed and achieved good results.In this theisi,these computational methods are divided into three categories:(1)lncRNA disease association prediction based on machine learning;(2)lncRNA disease association prediction based on biological network;(3)lncRNA disease association prediction based on deep learning.Based the analysis of lncRNA disease association prediction methods,we propose two excellent association prediction models.The main contents of this thesis are described below:(1)Considering the problems of isolated nodes and low model performance in lncRNA disease association prediction algorithms,this paper proposes a multi-view graph convolutional neural network-based lncRNA disease association prediction method,called MVGCN.The model fused multiple sources of information to improve the model’s ability to predict potential associations.The MVGCN model has two significant advantages:First,considering that lncRNA-miRNA-disease is an important regulation mechanism for lncRNA,the model uses regulatory data about miRNAs,and ablation experiments show the effectiveness of this approach.Second,using GCN and GAT to jointly extract features and fuse multiple features.Experiments show that MVGCN model has better performance than other models.(2)Considering the imbalance of positive and negative samples in lncRNA disease association prediction,we propose a lncRNA disease association prediction algorithm based on deep belief network and ensemble learning,called DBNDT.DBNDT model uses an ensemble learning method based on data resampling to solve the problem of imbalanced learning.The model firstly learns edge features based on DBN unsupervised learning nodes,removes data noise and reduces dimensionality,then uses multiple resampling datasets to train multiple classifiers,and finally uses a soft voting algorithm to merge the prediction results of multiple classifiers.Ablation experiments show that with the increase of the number of classifiers,the performance of the model is significantly improved,which indicates that the ensemble learning algorithm can effectively solve the problem of data imbalance.To verify the performance of the two models described in this paper,we evaluated the models using five-fold cross-validation,and the experiments showed that both models can effectively predict potential associations of lncRNA diseases.