Prediction Of CircRNA-disease Association Based On Graph Neural Network

Circ RNAs are circular endogenous non-coding RNAs with special structures that can be selectively spliced.In recent years,with the continuous progress of bioinformatics,researchers have found that many diseases are closely related to circ RNAs and play a very key role in the occurrence of diseases.Therefore,it is of great significance to explore the pathogenesis of circ RNAs and diseases and how to use circ RNAs to better prevent and treat diseases.However,using traditional biological experimental techniques to discover the interaction between new diseases and circ RNAs is very time-consuming and labor-intensive,so there is an urgent need to use advanced computational methods to study the association between diseases and circ RNAs more efficiently.The core idea of this thesis is to combine multi-kernel learning,conditional random fields and graph convolutional neural networks to extract deeper circ RNA-disease association features.First,the similarity between circ RNAs and diseases was calculated separately using the Gaussian interaction kernel function.Aiming at the similarity computation of diseases,this thesis also utilizes two-layer disease semantic similarity models respectively,and then fuses the three similarity methods as the final disease similarity.In this thesis,a three-layer graph convolution model is constructed,the conditional random field model is added to the second convolution layer to preserve the similarity characteristics in the graph embedding kernel learning,and a kernel operation is added to each convolutional layer,and the The final results of the three kernel operations were averaged and used to predict the association between diseases and circ RNAs through a distributed Laplace regularized least squares model.Finally,five-fold cross-validation is used to evaluate the performance of the model.According to the comparison of research results,the results obtained in this thesis under the Circ R2 Disease dataset are better than other models.In addition,this thesis also conducts empirical analysis on three diseases of lung cancer,pancreatic cancer and breast cancer.For the top 50 circ RNAs predicted by the model for these three diseases,39,43 and 37 have been confirmed in relevant databases,respectively.It shows that the predictive performance of the model is reliable.