基于多核学习和图卷积网络的药物-疾病关联预测  

Prediction of Drug-Disease Association Based on Multiple Kernel Learning and Graph Convolutional Networks

作　　者：陈书新 李玉田 王林 CHEN Shuxin;LI Yutian;WANG Lin(College of Artificial Intelligence,Tianjin University of Science and Technology,Tianjin 300457,China)

机构地区：[1]天津科技大学人工智能学院,天津300457

出　　处：《天津科技大学学报》2025年第1期64-71,共8页Journal of Tianjin University of Science & Technology

基　　金：天津市自然科学基金重点项目(20JCZDJC00140)。

摘　　要：识别和预测特定药物与疾病之间的关联关系,是药物研发过程中必不可少的一部分。现有方法对药物和疾病的多种异源信息整合不足。本文提出了一种基于多核学习和图卷积网络的计算方法预测药物-疾病关联。首先,对于药物相似度,基于药物-疾病关联矩阵和药物化学结构特征信息构建多个相似度核矩阵;同样,对于疾病相似度,基于关联矩阵构建多个相似度矩阵,并结合疾病语义相似度。其次,对这些相似度矩阵使用基于中心核对齐的多核学习算法进行整合。然后,构建基于图卷积网络的模型处理相似度网络和关联网络,从而提取药物和疾病特征。最后,使用内积解码器预测药物-疾病关联。与现有方法对比,本预测模型可以更准确地预测药物-疾病关联。Identifying and predicting the associations between specific drugs and diseases is an essential part of the drug development process.The previous computational methods did not well integrate the multiple heterogeneous information of drugs and diseases.In this article,a novel computational method based on multiple kernel learning and graph convolutional networks is proposed for drug-disease association prediction.Firstly,multiple similarity kernel matrices for drugs are constructed based on the association matrix and drug chemical structures.Similarly,multiple similarity matrices for diseases are constructed based on the association matrix,combined with disease semantic similarity.Secondly,these similarity matrices are integrated with the use of a center kernel alignment-based multiple kernel learning algorithm.A graph convolutional net-work model is then constructed to process the similarity network and association network,extracting features of drugs and diseases.Finally,an internal product decoder is used to predict drug-disease associations.In the experimental results,it was found that this model could predict the drug-disease associations more accurately than the state-of-the-art methods.

关 键 词：药物 疾病 药物-疾病关联 多核学习 图卷积网络 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]