DLGCN:基于图卷积网络的药物-lncRNA[]关联预测

为实现高通量识别新的药物-长链非编码RNA(Long non-coding RNA, lncRNA)关联,本文提出了一种基于图卷积网络模型来识别潜在药物-lncRNA关联的方法DLGCN(Drug-LncRNA graph convolution network)。首先,基于药物的结构信息和lncRNA的序列信息分别构建了药物-药物和lncRNA-lncRNA相似性网络,并整合实验证实的药物-lncRNA关联构建了药物-lncRNA异质性网络。然后,将注意力机制和图卷积运算应用于该网络中,学习药物和lncRNA的低维特征,基于整合的低维特征预测新的药物-lncRNA关联。通过效能评估,DLGCN的受试者工作特性曲线下面积(Area under receiver operating characteristic, AUROC)达到0.843 1,优于经典的机器学习方法和常见的深度学习方法。此外,DLGCN预测到姜黄素能够调控lncRNA MALAT1的表达,已被最近的研究证实。DLGCN能够有效预测药物-lncRNA关联,为肿瘤治疗新靶点的识别和抗癌药物的筛选提供了重要参考。

DLGCN: Prediction of drug-lncRNA associations based on graph convolution network

To realize high-throughput identification of new drug-lncRNA associations, we propose a new method DLGCN (Drug-LncRNA graph convolution network) to identify potential drug-lncRNA associations. First, we construct drug-drug and lncRNA-lncRNA similarity networks based on drug structure information and lncRNA sequence information, and then combine them with known drug-lncRNA associations to construct drug-lncRNA heterogeneous network. Next, the attention mechanism and graph convolution operation are applied to the network to learn the low dimensional features of drugs and lncRNAs. The new drug-lncRNA associations are predicted based on the integrated low dimensional features. DLGCN identified the drug-lncRNA associations with an AUROC (Area under the receiver operator characteristic) of 0.843 1, which is superior to classical machine learning methods and common deep learning methods. In addition, DLGCN predict that curcumin could regulate MALAT1, which has been confirmed by recent studies. DLGCN can effectively predict drug-lncRNA associations, which provides an important reference for identification of new tumor therapeutic targets and development of anti-cancer drugs.