Prediction of circRNA and Disease Association Based on GraphSAGE Model

环状RNA (circRNA)是一类内源性的非编码RNA，许多研究表明circRNA在复杂疾病中发挥着重要作用。然而，由于circRNA的功能复杂性和实验验证的高成本，传统的实验方法难以高效挖掘circRNA与疾病的关联关系，因此迫切需要高效的计算方法来揭示circRNA与疾病的关联关系。在现有数据库的基础上，本文提出了一种基于GraphSAGE模型的circRNA与疾病关联预测方法，通过整合circRNA相似性、疾病相似性以及已知的circRNA-disease关联数据构建异质图，随后借助GraphSAGE模型获得异质图中节点对应特征的高阶聚合表示，从而有效预测circRNA-disease关联。实验结果表明，GraphSAGE模型的AUC值为0.921，F1-score为0.865，Precision为0.879，Recall为0.852，以上四个评估指标均优于现有的DWNN-RLS和RWR模型。总之，GraphSAGE是预测circRNA-disease关联的有效方法。Circular RNA (circRNA) is a class of endogenous non-coding RNAs. Many studies have shown that circRNA plays an important role in complex diseases. However, due to the functional complexity of circRNA and the high cost of experimental verification, it is difficult for traditional experimental methods to efficiently mine the association between circRNA and disease, so efficient computational methods are urgently needed to reveal the association between circRNA and disease. Based on the existing database, this paper proposed a method for predicting the association between circRNA and disease based on GraphSAGE model. By integrating circRNA similarity, disease similarity and known circRNA-disease association data, a heterogeneous graph network was constructed, and then a high-level aggregated representation of the corresponding features of nodes in the heterogeneous graph network was obtained by GraphSAGE model, so as to effectively predict the circRNA-disease association. The experimental results demonstrate that the GraphSAGE model achieves an AUC of 0.921, F1-score of 0.865, Precision of 0.879 and Recall of 0.852, all of which were better than the existing DWNN-RLS and RWR models. In conclusion, GraphSAGE is an effective method to predict the association of circRNA-disease.