药物功能模式相似度及其聚类  

Drug Mode of Action Similarity and Clustering

作　　者：王志刚[1] 陈鑫[1] 谢丽芳 杨啸林[1] 张正国[1] 

机构地区：[1]中国医学科学院基础医学研究所北京协和医学院基础学院,北京100005 [2]中国医学科学院信息中心,北京100009

出　　处：《中国生物医学工程学报》2011年第6期807-812,共6页Chinese Journal of Biomedical Engineering

基　　金：国家科技支撑计划(2008BAI52B02);国家科技部科技基础性工作专项(2006FY110300);国家自然科学基金(30800241)

摘　　要：对现有已知药物的功能模式进行分析,可以帮助发现其可能的新应用,指导联合用药或预测药物的未知毒副作用。提出将药物化学结构信息和GO注释信息结合,分析药物功能模式相似度。药物化学结构和GO注释信息下载自DrugBank数据库,其中GO注释信息包括生物过程、分子功能和细胞定位等3个分支。计算现有4886种药物的功能模式相似度,并对其进行聚类分析。基于Tanimoto系数计算药物化学结构相似度,基于语义分析计算药物GO注释中3个分支的相似度。分别使用Logistic回归、算术均值、几何均值将上述4个药物相似度结合,得到反映多方信息的复合相似度。将一种药物与所有其他药物的相似度向量作为该药物的特征谱,对药物进行层次聚类。使用药物解剖学、治疗学及化学分类(ATC)的标准评价不同的相似度和聚类结果。结果显示:药物化学结构相似度与基于GO的3个分支的相似度均线性相关,表明药物的结构信息能在一定程度上反映功能信息;Logistic回归复合相似度能够很好地反映两个药物是否属于同一个ATC分类;基于GO注释生物过程分支语义相似度和几何均数复合相似度聚类结果与ATC分类第一层次强关联。所提出的方法结果可靠,可望用于辅助药物发现和预测不良反应。Comprehensive knowledge of the mode of action （MOA） of available drugs is crucial in their application expansion, combined administration, and side effect prediction. In this paper, drug chemical structure and three branches of GO annotation were combined to analyze the MOA similarities of the drugs. The chemical structure data and the annotation data of three branches of GO （biological processing, molecular function and cellular component） of the drugs were extracted from DrugBank database. The MOA similarities of 4 886 drugs were calculated and clustered. The chemical structure similarity was calculated with Tanimoto coefficient. GO annotation similarities for each branch were calculated by semantic analysis. Logistic regression, arithmetic mean and geometric mean were used separately to combine the four similarity results to form a composite similarity score. A similarity vector of one drug to all the others formed the feature spectrum of this drug. Hierarchical clustering was performed based on the feature spectrum of each drug. Anatomical Therapeutic Chemical （ATC） classification system was used to evaluate the similarity integration strategies and clustering results. The evaluation results indicated that the chemical structure similarity was correlated with the three GO-based similarities, suggesting that the drug structure information reflected the MOA. Logistic re^ression comoosite similarity was ~,enerallv in accordance with drum ATC classification. The clustering, resultsof BP and geometric mean composite similarities were strongly associated to the first level of ATC. The proposed similarity estimation and clustering methods can reveal the drug MOA and may be used in drug discovery and side effect prediction.

关 键 词：药物发现 功能模式 相似度 聚类 

分 类 号：R318[医药卫生—生物医学工程]