Zhang Qiaosheng, Sun Zhenyu, Zhong Zhaoman, Yang Huihui, Wei Yalong, Xu Junjie
School of Computer Engineering, Jiangsu Ocean University, No. 59, Cangwu Road, Haizhou District, Lianyungang, 222000, Jiangsu, China.
Jiangsu Institute of Marine Resources Development, Jiangsu Ocean University, No. 59, Cangwu Road, Haizhou District, Lianyungang, 222000, Jiangsu, China.
BMC Bioinformatics. 2025 Jul 15;26(1):177. doi: 10.1186/s12859-025-06195-0.
Accurately identifying drug-target interactions (DTI) can greatly shorten the drug development cycle and reduce the cost of drug development. In current deep learning-based DTI prediction models, the extraction of drug and target features is a key step to improve model performance. At the same time, drugs and targets form a complex relational network, and leveraging existing network topological relationships can obtain more comprehensive feature representations.
We propose a DTI prediction model based on sequence attribute extraction and graph neural networks, named SaeGraphDTI. First, sequence feature extractors are applied to extract relevant properties of drug and target sequences. Then, based on similarity relationships, the existing relational network is supplemented, and the graph encoder updates node information based on this network. Finally, the graph decoder calculates the probability of edge existence to predict DTI. Our model was compared with other state-of-the-art methods on four public datasets and achieved the best results in most key metrics.
These results demonstrate the excellent capability of this model in predicting potential DTI, providing a valuable tool for drug development.
准确识别药物-靶点相互作用（DTI）能够极大地缩短药物研发周期并降低药物研发成本。在当前基于深度学习的DTI预测模型中，药物和靶点特征的提取是提高模型性能的关键步骤。同时，药物和靶点构成一个复杂的关系网络，利用现有的网络拓扑关系可以获得更全面的特征表示。
我们提出了一种基于序列属性提取和图神经网络的DTI预测模型，名为SaeGraphDTI。首先，应用序列特征提取器来提取药物和靶点序列的相关属性。然后，基于相似性关系补充现有的关系网络，图编码器基于此网络更新节点信息。最后，图解码器计算边存在的概率以预测DTI。我们的模型在四个公共数据集上与其他现有最佳方法进行了比较，并在大多数关键指标上取得了最佳结果。
这些结果证明了该模型在预测潜在DTI方面的卓越能力，为药物研发提供了一个有价值的工具。
