首页|A Unified Graph Attention Network Based Framework for Inferring circRNA-Disease Associations
A Unified Graph Attention Network Based Framework for Inferring circRNA-Disease Associations
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
Springer Nature
Researchers have identified a large number of circular RNAs (circR-NAs). More and more studies have shown that circRNAs play crucial roles in regulating gene expression and function in distinct biological processes. CircRNAs are highly stable and conservative, which are suitable used as diagnostic biomark-ers for many human diseases. However, experimental verification of relationship between circRNAs and diseases is time-consuming and laborious, resulting in few known associations. The computational methods have subsequently been introduced to predict potential disease related circRNAs. Existing methods are not good enough in feature extraction and prediction performance. In this paper, we design a unified Graph Attention Network (GAT) framework to infer unknown circRNA-disease links. Our method unifies the feature extraction encoder and predictive decoder. To be specific, GAT based encoder with additional information are applied to learn representations of circRNAs and diseases. Then, several decoders, such as dot decoder, bilinear decoder, Neural Networks (NN) based decoder are implemented as predictors. Furthermore, we conduct detailed experimental analysis on the benchmark datasets based on 5-fold cross-validation. The evaluation metrics, including accuracy, precision, F1 score, AUC and AUPR values, and case studies of experimental results indicate that our method is efficient and robust for predicting circRNA-disease associations.
circRNADiseasecircRNA-disease association predictionGraph attention networks
NETL
Springer Nature
