基于残差结构的图卷积网络的药物靶点亲和力预测

Prediction of Drug-Target Affinity Based on Residual Structure Graph Convolutional Network

准确的药物靶点亲和力预测(DTA)能够缩短药物研发周期,节省人力和物力,加速药物研发过程。图神经网络(GNN)在药物靶点亲和力预测中得到了广泛应用,但现有的方法大多基于浅层GNN。该文提出了一种基于残差结构的图卷积网络,残差结构的加入能够加深网络结构,借此构建一个具有24个图卷积层的深度图卷积网络,以此捕获药物分子的特征,学习有效的嵌入表达,并在两个基准药物靶点亲和力数据集上与几种先进的基于机器学习或深度学习的模型进行比较。结果表明,该文所提模型相较于其他基准模型有着更好的预测性能,验证了该文所提方法的有效性。

Accurate drug target affinity prediction(DTA)can shorten the drug development cycle,save manpower and material resources,and accelerate the drug development process.Graph Neural Networks(GNN)have been widely used in drug target affinity prediction,but most of the existing methods are based on shallow GNN.Therefore,a graph convolutional network based on the residual structure is proposed.The addition of the residual structure can deepen the network structure,thereby constructing a deep graph convolutional network with 24 graph convolutional layers to capture the characteristics of drug molecules,learn efficient embedding representations,and compare with several state-of-the-art machine learning or deep learning based models on two benchmark drug target affinity datasets.The results show that the proposed model has better predictive performance than other benchmark models,which verifies the effectiveness of the method proposed in this paper.