School of Information Engineering, East China Jiaotong University, Nanchang, China.
College of Computer Science and Electronic Engineering, Hunan University, Changsha, China.
J Biomed Inform. 2020 Dec;112:103624. doi: 10.1016/j.jbi.2020.103624. Epub 2020 Nov 18.
A growing body of experimental studies have reported that circular RNAs (circRNAs) are of interest in pathogenicity mechanism research and are becoming new diagnostic biomarkers. As experimental techniques for identifying disease-circRNA interactions are costly and laborious, some computational predictors have been advanced on the basis of the integration of biological features about circRNAs and diseases. However, the existing circRNA-disease relationships are not well exploited. To solve this issue, a novel method named DeepWalk and network consistency projection for circRNA-disease association prediction (DWNCPCDA) is proposed. Specifically, our method first reveals features of nodes learned by the deep learning method DeepWalk based on known circRNA-disease associations to calculate circRNA-circRNA similarity and disease-disease similarity, and then these two similarity networks are further employed to feed to the network consistency projection method to predict unobserved circRNA-disease interactions. As a result, DWNCPCDA shows high-accuracy performances for disease-circRNA interaction prediction: an AUC of 0.9647 with leave-one-out cross validation and an average AUC of 0.9599 with five-fold cross validation. We further perform case studies to prioritize latent circRNAs related to complex human diseases. Overall, this proposed method is able to provide a promising solution for disease-circRNA interaction prediction, and is capable of enhancing existing similarity-based prediction methods.
越来越多的实验研究表明，环状 RNA（circRNAs）在致病机制研究中很有意义，并且正在成为新的诊断生物标志物。由于鉴定疾病相关 circRNA 的实验技术既昂贵又费力，因此一些基于 circRNA 和疾病的生物学特征整合的计算预测器已经得到了发展。然而，现有的 circRNA-疾病关系并没有得到很好的利用。为了解决这个问题，提出了一种名为 DeepWalk 和网络一致性投影的用于 circRNA-疾病关联预测的新方法（DWNCPCDA）。具体来说，我们的方法首先揭示了基于已知 circRNA-疾病关联的深度学习方法 DeepWalk 学习的节点特征，以计算 circRNA-circRNA 相似性和疾病-疾病相似性，然后将这两个相似性网络进一步用于网络一致性投影方法，以预测未观察到的 circRNA-疾病相互作用。结果表明，DWNCPCDA 对疾病 circRNA 相互作用预测具有高精度性能：留一交叉验证的 AUC 为 0.9647，五折交叉验证的平均 AUC 为 0.9599。我们进一步进行案例研究，以优先考虑与复杂人类疾病相关的潜在 circRNAs。总的来说，该方法为疾病 circRNA 相互作用预测提供了一种有前途的解决方案，并能够增强现有的基于相似性的预测方法。
