MVGNCDA: Identifying Potential circRNA-Disease Associations Based on Multi-view Graph Convolutional Networks and Network Embeddings.

Increasing evidences have indicated that circular RNAs play a crucial role in the onset and progression of various diseases. However, exploring potential disease-associated circRNAs using conventional experimental techniques remains both time-intensive and costly. Recently, various computational approaches have been developed to detect the circRNA-disease associations. Nevertheless, due to the sparsity of the data and the inefficient utilization of similarity representation, it is still a challenge to effectively detect unknown circRNA-disease associations using multisource data. In this work, we propose an innovative computational framework, MVGNCDA, which merges a multi-view graph convolutional network (GCN) and biased random walk-based network embeddings to evaluate potential circRNA-disease associations from multisource data. First, we calculate disease semantic similarity, circRNA functional similarity, and their Gaussian interaction profile (GIP) kernel and cosine similarity. MVGNCDA utilizes multi-view GCNs to extract local node embeddings of diseases and circRNAs in the context of multisource information. Then, we construct a heterogeneous network utilizing integrated similarity and verified circRNA-disease associations, which is subsequently used to learn global node embeddings. Furthermore, the final fused local and global node embeddings are decoded to evaluate the circRNA-disease associations using a bilinear decoder. The fivefold cross-validation results demonstrate that MVGNCDA outperforms existing methods across five public datasets. Moreover, case study also confirms that MVGNCDA is capable of efficiently identifying unknown circRNA-disease associations.