SGCNCMI: A New Model Combining Multi-Modal Information to Predict circRNA-Related miRNAs, Diseases and Genes

SIMPLE SUMMARY: With the development of circRNA–miRNA-mediated models, circRNAs have been shown to play a prominent role in the development and treatment of diseases such as cancer, and unearthing potential miRNA-associated circRNAs may provide new insights and ideas for the diagnosis and treatment of complex diseases such as cancer. Large-scale prediction using computer technology can provide an a priori guide to biological experiments and save costs. This paper presents the third computational method in this field with the highest accuracy to date, and we also collected and integrated high-quality datasets from the current database, which we believe will allow future computational innovations to develop. ABSTRACT: Computational prediction of miRNAs, diseases, and genes associated with circRNAs has important implications for circRNA research, as well as provides a reference for wet experiments to save costs and time. In this study, SGCNCMI, a computational model combining multimodal information and graph convolutional neural networks, combines node similarity to form node information and then predicts associated nodes using GCN with a distributive contribution mechanism. The model can be used not only to predict the molecular level of circRNA–miRNA interactions but also to predict circRNA–cancer and circRNA–gene associations. The AUCs of circRNA—miRNA, circRNA–disease, and circRNA–gene associations in the five-fold cross-validation experiment of SGCNCMI is 89.42%, 84.18%, and 82.44%, respectively. SGCNCMI is one of the few models in this field and achieved the best results. In addition, in our case study, six of the top ten relationship pairs with the highest prediction scores were verified in PubMed.