【摘要】 使用ＣＡＳＡＣ软件中的逐步回归和改进的神经网络（ＡＮＮ）方法，对Ｎ．Ｏｉｋａｗａ等研究合成的杀幼（虫）剂进行了计算．其中，４４个化合物作为训练基，１１个化合物作为未知样本，获得良好的预报结果，与Ｎ．Ｏｉｋａｗａ等使用Ｈａｎｓｃｈ方法计算所取得的结论一致．不同的是本文所使用的物化参数除Ｂ５和Ｌ１之外，都能够方便的计算．改进的神经网络（ＡＮＮ）方法提高了模型质量和预报结果的精度．更多还原

【Abstract】 A new technique has been proposed for molecular design,which is a sequential combination of the stepwise regression and improved artificial neural networks.Unlike the traditional method, the process of numerization of structural information is fully automatic.All Physicochemical parameters in the model are generated by means of both the parameter database and structural matching module in CASAC software.Specially,the improved artificial neural networks,with rational convergence criterion to avoid overfitting,improved model’s quality and expanded the scope of application to nonlinear systems.With 55 larvicidal 1-(Substituted benzoyl)-2-benzoyl-1-tert-butylhydrazine compounds synthesized by N.Oikawa etc.,44 of them are taken as the training set and the rest eleven as testing set, the proposed method gave excellent result.For bio-activity prediction of the 11 testing compounds,the average error is 0.33.The proposed method can be used as a convenient routine technique in molecular design.更多还原