神经网络法在喹诺酮类化合物定量构效关系中的应用

Application of Neural Network Method on the Quantitive Structure-Activity Relationship of Quinolones Compounds

目的:以神经网络法研究喹诺酮类化合物定量构效关系。方法:利用Matlab软件包构建一个3层BP神经网络,对数据集进行计算,并将结果与线性回归法的计算结果进行比较。结果:神经网络法的误差平方和为0 .3042 ,小于线性回归法,预测的相关系数为0. 86。结论:神经网络法在喹诺酮类化合物定量构效关系研究中获得了比回归法更精密的拟合结果。

OBJECTIVE:To study the quantitative structure-activity relationship of quinolones compounds by neural network(NN)method.METHODS:A3-layered BP neural network was constructed with the Matlab software package,the collected data were calculated,result of which was compared with that of the linear regression.RESULTS:Sum of square of errors for the neural network method was0.3042,which was less than that of linear regression;the predicted correlation co-efficient was0.86.CONCLUSION:The neural network method has achieved more precise fitting results than the linear re-gression in the study of the quantitative structure-activity relationship of quinolones compounds.