Tensor product graph diffusion based on nonlinear fusion of multi-source information to predict circRNA-disease associations

Abstract

Circular RNA (circRNA), a non-coding RNA, has been shown in recent years via numerous studies to be more than just a "by-product" of gene transcription. It also participates in the regulation of many diseases and influences various biological functions in the human body. Its unique properties and biological functions indicate that circRNA may become new biomarker and therapeutic target, thus, it is crucial for modern medicine to identify potentially relevant circRNAs. This paper proposed novel computational model SNFTPGd-CDA for circRNA-disease associations prediction with the CircR2Disease v2.0 as dataset, which first constructs multiple similarity networks according to multi-source information related to circRNA and disease, and uses similarity network fusion (SNF) to nonlinearly fuse multi-source similarity networks. Then tensor product graph diffusion (TPGd) is employed to diffuse the similarity in the tensor product graph of the fused similarity network into the context of other nodes, which in turn constructs a more reliable similarity network as the feature matrix. Finally, the cascaded forest model is employed for circRNA-disease associations prediction. An AUC value of 0.9937 and an accuracy of 96.92 % were obtained with five-fold cross-validation. This result indicates that SNFTPGd-CDA achieves more accurate performance than previous methods, and can support validation of circRNA-disease associations.