神经网络用于环丙胺类衍生物的构效关系研究

NEURAL NETWORKS IN QSAR STUDIES:ESTIMATION AND PREDICTION OF BIOLOGICAL ACTIVITY FOR N-(SUBSTITUTED PHENOXYETHYL)=CYCLOPROPYLAMINES

将神经网络应用于定量构效关系研究。用改进的反传算法探讨了单胺氧化酶抑制剂Ｎ－（苯氧乙基）环丙胺取代衍生物的生物活性与取代基电子效应σ、疏水作用π、空间效应Ｅｓ等参数之间的定量关系。给出了精密拟合和准确预测（最大误差均小于１０％），优于经典的多元线性回归及逐步回归方法。作为一种有效的计量化学新方法，神经网络有良好的预测能力和非线性处理功能，从而可望在ＱＳＡＲ研究中发挥重要作用。

Neural networks(NN)methods were applied to quantitative structure-activity relationship(QSAR)studies.The relationship between biological activity(pC=pIC50)and electronic,hydrophobic and steric parameters and dumming index(sigma,pi,Es,was investigated by usingmodified backpropagation(MBP)neural networks.The biological activity of N-(substitutedphenoxyethyl)-cyclopropylamines derivative regression was estimated and predicted with relative errorless than 16%and with correct classification ratio being94.4%,The results obtained by thedeveloped NN(MBP)method seem to be better than those by multivariate regression(MR)and step-wise regression(SR).The NN(MBP)method was,therefore,regarded as an excellent chemometricmedeling technique for estimating and predicting biological activity on basis of chemical structure forQSAR studies。