人工神经网络改进及其在喜树碱类化合物定量构效关系研究中的应用

Modification of artificial neural networks and its application in the Q SAR studies of antitumoral camptothecin derivatives

目的 采用神经网络方法对喜树碱类抗肿瘤药物构效关系进行考察。方法 采用主成分分析 (principalcomponentanal ysis,PCA)预处理输入层数据和Levenberg Marquardt规则训练网络。 结果 改进的神经网络PCANN所得结果预测性能较强(R2 值为 0 .878) ,明显优于GA PLS结果 (R2 值为 0 .684)。结论 本研究表明经过改进后的神经网络不仅训练速度大大提高 ,而且所得结果统计意义显著。这为下一步合理设计。

OBJECTIVE: Artificial neural networks (ANNs) were applied into the QSARs of 21 antitumor camptothecin derivatives. METHOD: Principal component analysis (PCA) was used to preprocess training data set to reduce the dimension of input vectors and orthogonalize the components of input vectors. Artificial neural networks were trained with Levenberg-Marquardt rule that has been proved to have a fastest convergence for modest network. RESULTS: The results obtained by the PCANN method (R2 = 0.878) seemed to be much better than those by GA-PLS (R2 = 0.684). CONCLUSION: It was shown that modified artificial neural network - PCANN (Principal Component Analysis/Artificial Neural Network) had a faster convergence and the resulting model had good predictive ability. This will benefit future design and synthesis of novel highly potent antitumor camptothecin derivatives.