GCNMFCDA: A Method Based on Graph Convolutional Network and Matrix Factorization for Predicting circRNA-Disease Associations

Numerous studies reveal that Circular RNAs (circRNAs) are critical for human physiological and pathological processes. Research on the disease-related circRNAs can provide insight into the mechanisms of extraordinary disease and benefit the prevention and therapy of numerous untreated human diseases. As biological experiments require a great deal of labor and time, the calculation-based methods have become more and more salient in the field. Here we present a brand-new model named GCNMFCDA, which contains Graph Convolutional Network (GCN) and Matrix Factorization (MF) to predict disease-related circRNAs. Theoretically, we first utilize circRNA and disease similarities to construct initial node features. Then we apply GCN to learn potential embeddings of circRNA and disease based on initial node features. Finally, the inner product between circRNA embedding and disease embedding is introduced to construct the new score matrix based on matrix factorization. We adopt fivefold and tenfold cross validation to evaluate our model. GCNMFCDA obtains an average AUC of 0.9330 and 0.9290, respectively, and performs better than the other eleven existing methods. Furthermore, case studies on breast cancer and glioma illustrate that 19 and 18 of the top 20 candidate circRNAs were respectively confirmed in the validation datasets or published literature. These experimental results reveal that GCNMFCDA can discover circRNA-disease associations effectively and reliably.