Prediction Of Circrna-disease Association Based On Graph Learning

CircRNA is a kind of non-codingRNA without 5 and 3polyadenylated tails.It is different from linearRNA that the circRNA is relatively stable as it can escape the digestion of exonuclease.With the development of the high-through sequencing technologies,accumulating studies have demonstrated that circRNA participates in lots of biological activities and has a close relationship with human diseases,such as Alzheimer’s disease.diabetes,heart disease,cancer.Predicting the circRNA-disease associations can help to learn the pathogenesis of the disease and further for diagnosis and treatment.Many circRNA-disease associations have been found by using biological methods.However,it is time-consuming and labor-consuming.In recent years,plenty of predicting circRNA-disease associations methods have been proposed.The performance of many methods is limited due to the sparse associations between circRNAs and diseases.How to address the problem of sparse circRNA-disease associations and how to improve the performance of the methods with multiple biological data remain further research.In order to address the above problems,two computational methods based on graph network learning are proposed to predict circRNA-disease associations.The embedding of the circRNA and disease will contain multiple neighbor information by using graph neural network and knowledge graph in the heterogenous network.Specifically,the two major works are as follows:(1)In this thesis,we present a new model(IGNSCDA)based on Improved Graph convolutional network and Negative Sampling to predict CircRNA-Disease Associations.In our method,it first constructs the heterogeneous network on the basis of the known circRNA-disease associations.Then,an improved graph convolutional network is designed to obtain the feature vectors of circRNA and disease.Further,the multilayer perceptron is employed to predict circRNA-disease associations based on the feature vectors of circRNA and disease.In addition,the negative sampling method is used to reduce the effect of the noise samples,which selects negative samples based on the expression profile similarity and Gaussian Interaction Profile kernel similarity of circRNAs.The results of 5-fold cross validation show that IGNSCDA outperforms than other state-of-the-art methods in the prediction performance.Moreover,the case study shows that IGNSCDA is effective for identifying the circRNAs associated with diseases.The code of IGNSCDA and the circRNA-disease associations are released at https://github.com/lanbiolab/IGNSCDA for further research.(2)In this thesis,we propose a new algorithm(KGANCDA)to identify circRNA-disease associations based on knowledge graph attention network.The circRNA-disease knowledge graphs are constructed by collecting multiple relationship data among circRNA,disease,miRNA and lncRNA.Then,the knowledge graph attention network is designed to obtain embeddings of each entity by distinguishing the importance of information from neighbors.Besides the low-order neighbor information,it can also capture high-order neighbor information from multi-source associations,which alleviates the problem of data sparsity.Finally,the multi-layer perceptron is applied to calculate the affinity score of circRNAdisease associations based on the embeddings of circRNA and disease.The experiment results demonstrate that KGANCDA outperforms than other state-of-the-art methods in 5-fold cross validation.Furthermore,the case study demonstrates that KGANCDA is effective to identify potential circRNA-disease associations.The code of KGANCDA and the two knowledge graphs(cancer and non-cancer knowledge graphs)are released at https://github.com/lanbiolab/KGANCDA for further research.