CircRNA-disease inference using deep ensemble model based on triple association

Abstract

Accumulating evidence indicates more and more circular RNAs (i.e. circRNAs) have played a vital role in regulating gene expression and are related to diseases through different biological procedures. Predicting circRNA-disease associations helps to conjecture possible disease related circRNA and facilitate human disease diagnosis and downstream treatment. Nevertheless, little effort was made to uncover the interaction between various diseases and circRNAs. In our work, human circRNA-disease association network is first generated using known miRNA-circRNA interactions and disease related miRNA (microRNA) information. Then we further integrated this information to compute similarity scores between human diseases and circRNAs. Here, we proposed one deep ensemble model called DeepInteract, which first used two stacked auto-encoders to explore hidden features utilizing similarity information, and adopted a 3-layer neuron network to predict the final association. Our method is capable of capturing more complex non-linear features comparing to other approaches. Our results indicate the proposed method is superior to other previous competitors. Many prediction results have been validated by some biological experiments using our model.