An in-silico method with graph-based multi-label learning for large-scale prediction of circRNA-disease associations

Circular RNAs (circRNAs) have been proved to be implicated in various pathological processes and play vital roles in tumors. Increasing evidence has shown that circRNAs can serve as an important class of regulators, which have great potential to become a new type of biomarkers for tumor diagnosis and treatment. However, their biological functions remain largely unknown, and it is costly and tremendously laborious to investigate the molecular mechanisms of circRNAs in human diseases based on conventional wet-lab experiments. The emergence and rapid growth of genomics data sources has provided new opportunities for us to decipher the underlying relationships between circRNAs and diseases by computational models. Therefore, it is appealing to develop powerful computational models to discover potential disease-associated circRNAs. Here, we develop an in-silico method with graph-based multi-label learning for large-scale of prediction potential circRNA-disease associations and discovery of those most promising disease circRNAs. By fully exploiting different characteristics of circRNA space and disease space and maintaining the data local geometric structures, the graph regularization and mixed-norm constraint terms are also incorporated into the model to help to make prediction. Results and case studies show that the proposed method outperforms other models and could effectively infer potential associations with high accuracy.