多位取代喹诺酮药物的人工神经网络构效关系研究  

SAR study of multi-site substituted fluoroquinolones using artifical neural net

作　　者：刘浩[1] 屈凌波[2] 相秉仁[3] 

机构地区：[1]广东食品药品职业学院药学系,广东广州510520 [2]郑州大学化学系,河南郑州450052 [3]中国药科大学分析测试中心,江苏南京210009

出　　处：《计算机与应用化学》2007年第11期1479-1483,共5页Computers and Applied Chemistry

基　　金：河南省杰出人才创新基金资助(421002300);药职院院内自然课题基金(2006019)

摘　　要：利用人工神经网络研究69个喹诺酮类药物的构效关系,分别使用BP(Back Propagate)、LM(Levenberg-Marquardt)和RBF(径向基,Radial Basis Function)三种网络方法研究量子化学方法(MNDO法和AM1法)计算所得到的参数。BP网络结果不收敛,RBF网络收敛速度最快,不超过100次迭代就可得到较精确的迭代结果。LM和RBF两种网络预测训练集的正确率均达到100%。其中使用RBF网络和AM1法所得到的参数预测23个预测集的正确率82.6%,MNDO法预测预测集的正确率78.3%,体现了喹诺酮药物的氢键作用成分,结合多元回归分析和相关的耐药机理研究,提出喹诺酮药物的双氢键作用机理。人工神经网络方法可作为研究药物构效关系和研发新药物的有力工具。The SAR Study of Multi-site Substituted Fluoroquinolones was performed by Artifical Neural Net method, the method that used includes BP（Back propagate） , LM（LM algorithms） and RBF（Radial Basis Function） neural net. The parameters are computed by MNDO method and AM1 method. The RBF method has the best speed that it can get the training aim no more than 100 iteration. The forecast accuracy rate is 100% to the training set for the LM and RBF net. As for the test sets, the accuracy rate of AM1 method is 82.6% ,while MNDO method is 78.3%. AM1 method is a little better than MNDO method which is also relational with that AM1 method has hydrogen bond property. Integrating the related biological research of fluoroquinolones resistance, the binary hydrogen bond mechanism of fluoroquinolones is brought forward. It implies that fluoroquinolones realize their function as electron donors as well as a electron acceptors in its function process. The results of this paper can provide material for Pharmacody study, at the same time, it can be served as a method and a tool in the development of new medicine.

关 键 词：人工神经网络(ANN) 喹诺酮 构效关系(SAR) MNDO(改进的忽略微分重叠) AM1(Austin模型) 

分 类 号：TQ224.23[化学工程—有机化工]