摘要
将Vapnik提出的支持向量机(support vector machine,简称SVM)算法用于化合物活性的模式识别研究。SVM算法是特别适合于用有限已知样本训练建模,进而预报未知样本属性的模式识别新算法,将其用于N-(3-氧-3,4-双氢-2氢-苯并[1,4]噁嗪-6-羰基)胍类化合物的Na/H交换抑制活性类别的识别研究,用留一法考察了SVM模型的预报能力,并与Fisher判别矢量法和最近邻(KNN)法的分类预报结果进行比较,结果表明:SVM算法的预报结果优于Fisher法和KNN法的结果。因此,SVM算法可望应用于药物的构效关系研究领域。
Support vector machine proposed by Vapnik is a newly developed technique for data mining. It is suitable for the data processing based on finite nurnber of training samples, with special technique to restrict overfitting. In this work, the support vector classification (SVC) method was applied to structure activity relationship (SAR) study on N-(3-Oxo-3 ,4-dihydro-2H-benzo[1 ,4]oxazine-6-carbonyl) guanidines as Na/H exchange inhibitory agent. The modeling result obtained by SVC was better than that of Fisher method and KNN method. It is expected that the SVM method would be further applied to the field of QSAR studies.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2002年第6期745-748,共4页
Computers and Applied Chemistry
基金
由国家自然科学基金和上海宝钢集团公司联合资助(50174038)
