摘要
试用新近提出的、特别适合于小样本多变量训练集的支持向量机(support vector machine,简称SVM)算法于复杂药物分子设计。对一批26个处理化疗或放疗呕吐拮抗药的候选化合物筛选数据用留一法判别SVM的预报能力。结果表明:与人工神经网络、最近邻法(KNN)、Fisher法相比,SVM算法可以提供误报率更低的数学模型。
The relationship between the activity and structural descriptors of antagonists was investigated by using the support vector machine developed by Vapnik. For the sample set with 26 compounds as antagonists, the cross validation by leaving-one method was used to compare the prediction ability of support vector machine method with KNN and Fisher method. It was found that the prediction result by support vector machine was better than that of KNN or Fisher method.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2002年第6期741-744,共4页
Computers and Applied Chemistry
基金
国家自然科学基金委和美国福特公司联合资助(9716214)
