论文部分内容阅读
本文针对44个黄酮化合物对醛糖脱氢酶,采用贝叶斯正则化反向传播神经网络构建定量构效关系模型。选取116种与结构相关的分子描述符,通过遗传算法进行变量筛选。建立基于8个变量的活性预测贝叶斯正则化神经网络模型并采用验证集考察其预测性能。在该模型下,黄酮化合物对醛糖脱氢酶抑制活性的实验值和预测值一元相关系数平方(R2)分别为0.94811和0.97789。模型显示黄酮化合物醛糖脱氢酶抑制活性与其结构有密切关系。贝叶斯规整化神经网络结合遗传算法具有良好的预测能力。
In this paper, 44 flavonoid aldose dehydrogenase enzymes, using Bayesian regularized backpropagation neural network to construct quantitative structure-activity relationship model. 116 structure-related molecular descriptors were selected, and the variables were screened by genetic algorithm. An active predictive Bayesian regularization neural network model based on eight variables was established and its predictive performance was examined by using the verification set. Under this model, the univariate correlation coefficients (r 2) of the experimental and predicted values of flavonoid compounds for aldose dehydrogenase inhibition were 0.94811 and 0.97789, respectively. The model shows that the flavonoid aldose dehydrogenase inhibitory activity has a close relationship with its structure. Bayesian regularized neural network combined with genetic algorithm has good predictive ability.