论文部分内容阅读
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 structure-activity relationship of 69 quinolones was studied by using artificial neural network. Three network methods, BP (Back Propagate), LM (Levenberg-Marquardt) and RBF (Radial Basis Function) France and AM1 method) to calculate the resulting parameters. The results of BP network do not converge, RBF network converges fastest, and more than 100 iterations can get more accurate iteration results. LM and RBF network prediction training set both correct rate of 100%. Among them, the prediction accuracy of 23 prediction sets was 82.6% using the parameters obtained from RBF network and AM1 method, 78.3% from MNDO method predicted the prediction set, which reflected the hydrogen bonding effect of quinolone drugs. Combined with multiple regression analysis and related Resistant mechanism of quinolone drugs proposed double hydrogen bonding mechanism. Artificial neural network method can be used as a powerful tool to study the structure-activity relationship of drugs and to develop new drugs.