基于图神经网络的中药聚类方法研究  

Research on Clustering Method for Chinese Herbal Medicine Based on Graph Neural Network

作　　者：何佳怡 谢佳东 胡晨骏[1] 胡孔法[1,3] HE Jiayi;XIE Jiadong;HU Chenjun;HU Kongfa(School of Artificial Intelligence and Information Technology,Nanjing University of Chinese Medicine,Nanjing 210023,China;Institute of Literature in Chinese Medicine,Nanjing University of Chinese Medicine,Nanjing 210023,China;Jiangsu Collaborative Innovation Center of Traditional Chinese Medicine in Prevention and Treatment of Tumor,Nanjing 210023,China)

机构地区：[1]南京中医药大学人工智能与信息技术学院,南京210023 [2]南京中医药大学中医药文献研究所,南京210023 [3]江苏省中医药防治肿瘤协同创新中心,南京210023

出　　处：《世界科学技术-中医药现代化》2024年第11期2988-2995,共8页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：国家自然科学基金委员会面上基金项目(82074580):基于知识图谱的现代名老中医诊治肺癌用药规律及其机制研究,负责人:胡孔法;国家科学技术部国家重点研发计划重点专项课题(2022YFC3502302):全身望诊形态信息智能采集分析关键技术与装备,负责人:胡孔法;江苏省教育厅江苏省研究生科研创新计划(KYCX22_2054):基于知识图谱的名老中医诊治肺癌的用药机制研究,负责人:何佳怡。

摘　　要：目的本研究提出了一种基于图神经网络的中药聚类方法(CHM-GCNK),旨在从生物分子网络层面发现潜在的中药配伍。方法首先,收集中药、靶点(蛋白质)信息以及它们之间的相互作用关系,构建中药靶点网络。其次,采用图神经网络学习所构建的中药靶点网络,获取中药节点的嵌入表示。然后,利用Kmeans算法进行聚类。最后,采用非线性降维技术t-SNE可视化聚类结果。结果应用CHM-GCNK、Node2Vec-Kmeans和SVD-Kmeans方法,以治疗肺癌的40个中药为例进行聚类,聚类结果为五个簇,聚类算法评价指标SS、DBI、CH结果显示CHM-GCNK优于其他两种方法,分别为0.4006、0.7631、59.0001。结论CHM-GCNK聚类效果更好,可应用于中药配伍研究,进而为人工智能和多组学数据时代的中医药生物网络分析方法提供参考借鉴。Objective This study proposes a Chinese Herbal Medicine(CHM)clustering method based on graph neural network(CHM-GCNK),aiming to discover potential compatibility of CHM at the biological network level.Methods Firstly,collect data of CHM,target,and their interactions,and construct a network of CHM and targets.Secondly,the graph neural network is used to learn the constructed network and obtain the embedded representation of CHM nodes.Then,use the Kmeans algorithm to clustering.Finally,use nonlinear dimensionality reduction technology t-SNE to visualize clustering results.Results The CHM-GCNK,Node2Vec-Kmeans,and SVD-Kmeans methods were applied to cluster 40 CHM for the treatment of lung cancer.The clustering results were five clusters,and CHM-GCNK was superior to the other two methods.The evaluation indicators SS,DBI,and CH showed results of 0.4006,0.7631,and 59.0001,respectively.Conclusion The clustering effect of CHM-GCNK is better and can be applied to the study of CHM compatibility,providing reference for the analysis methods of CHM biological networks in the era of artificial intelligence and multi omics data.

关 键 词：图神经网络 蛋白质互作网络 分子生物学 中药聚类 

分 类 号：R-058[医药卫生]