学术
头孢菌素类抗生素的定量构效关系研究
王春娟[1] 谭显胜[2] 袁哲明[1] 熊洁仪[1]
[1]湖南农业大学生物安全科学技术学院,长沙410128 [2]湖南农业大学农学院,长沙410128
摘 要:
目的：建立一种预测精度较高的定量构效关系（QSAR）模型，为设计和合成活性更高的头孢菌素类抗生素提供理论依据。方法：发展了一种基于支持向量回归（SVR）和k-最近邻（KNN）的非线性组合预测方法（SVR-KNN），系统研究了48种抗流感嗜血杆菌头孢菌素衍生物的QSAR。结果：留一法预测结果表明，非线性筛选描述符和子模型能明显提高预测精度，汰选子模型后的组合预测精度优于单一子模型，SVR．KNN的MSE、MAPE分别为0．019、1．81％；独立样本预测结果显示，SVR-KNN在所有参比模型中具有最优的预测精度及稳定性，其MSE、MAPE分别为0．010、1．33％。结论：SVR-KNN模型具有较强的预测能力和优异的泛化推广能力，在抗生素及其他药物的QSAR研究中有广泛应用前景。[著者文摘]
文章出处:
《现代生物医学进展》-2007年7卷11期 -1718-1722页
栏目信息:
分 类 号:
文献标识码:
A
文章编号:
1673-6273（2007）11-1718-05
The Quantitative Structure-Activity Relationship of Cephalosporin Antibiotics
WANG Chun-juan, TAN Xian-sheng, YUAN Zhe-ming, XIONG Jie-yi （1 Bio-safety Science and Technology College, Hunan Agricultural University, Changsha 410128, China; 2 Agronomy College, Hunan Agricultural University, Changsha 410128, China）
Abstract:
Objective： To establish a quantitative structure-activity relationship （QSAR） model with higher prediction precision, and to provide theoretical data for the design of highly effective cephalosporin antibiotics. Methods： Based on support vector regression （SVR） and k-nearest neighbor （KNN）, a nonlinear combinatorial prediction approach, named SVR-KNN, was developed and applied to the QSAR on the antibacterial bioactivities of 48 cephalosporin compounds against Haemophilus influenzae. The approach consists of 6 steps. Firstly, search for the optimal SVR kernel automatically based on the minimal mean square error （MSE） in which leave-one-out method is used. Secondly, screen the descriptors nonlinearly. Thirdly, select training samples using KNN models, then predict based on SVR, and construct sub-models using the SVR predicted values of different k values training sets. Fourthly, get the optimal kernel based on sub-models. Then screen the sub-models nonlinearly. Finally, perform nonlinear combinatorial prediction using SVR by leave-one-out based on the retained sub-models. Results： The results of leave-one-out method showed that the nonlinearly screening of descriptors and sub-models was essential and combinatorial prediction after screening sub-models was more precise than the single model. The result of independent samples test also indicated that SVR-KNN had the highest prediction precision and stability of all reference models built from the same dataset. Conclusions： SVR-KNN has strong prediction ability and outstanding generalization ability. It is expected to be widely used in QSAR of antibiotics and other drugs.[著者文摘]
Key words:
Cephalosporin antibiotics; Quantitative structure-activity relationship; Support vector regression; K-nearest neighbor; Combinatorial prediction
基金资助:
国家自然科学基金资助项目（No.30570351）和教育部新世纪优秀人才计划资助项目
更多>>免费文章
cqvip.com