HCGCCDA: Prediction of circRNA-disease associations based on the combination of hypergraph convolution and graph convolution

With the advancement of modern biological experiments, an increasing body of evidence indicates a complex relationship between circRNA and diseases, which holds significant implications for disease research and treatment. However, traditional biological experiments have shown low efficiency and high costs when studying the association between circRNA and diseases. To address this issue, researchers have proposed various methods for predicting the correlation between circRNA and diseases. Nevertheless, existing methods mainly focus on feature extraction from individual nodes, leaving the feature extraction from edge information largely unexplored. In this research, we use an advanced method that performs information extraction from both edges and nodes for edge prediction. Specifically, we construct similarity networks for circRNA and diseases from multiple perspectives, including circRNA and disease feature networks obtained through matrix factorization, as well as similarity networks based on Pearson correlation and Gaussian kernel. We also leverage circRNA sequences and disease semantic information to construct similarity networks. Subsequently, we refine node feature information by extracting and optimizing features from both the nodes themselves and their neighbors. Next, we employ hypergraph convolutional networks and graph convolutional networks to learn and embed features from both dynamic and static aspects. Finally, we score the obtained comprehensive embedding features and utilize the model's training results for prediction. Our model demonstrates excellent performance in five-fold cross-validation, achieving an AUC value of 0.9833, surpassing other advanced models. The results of the case study further validate its high precision. Consequently, our model provides more valuable foundational information for biological research, with the potential to make significant impacts in disease research and treatment.