论文部分内容阅读
采用电性拓扑状态指数(En)表征昆虫酚氧化酶(PO)抑制剂的分子结构,通过最佳变量子集回归的方法建立了57种PO抑制剂抑制活性(pIC50)的多元线性回归模型,非交叉相关系数和交叉相关系数分别为0.920和0.908,经Jackknife和变异膨胀因子(VIF)检验具有良好的稳定性和预测能力.该模型显示影响PO抑制剂抑制活性的主要因素是—OH,—O—和C=O等分子结构片段.以模型中的3个参数E13,E14,E16为人工神经网络输入层,设定3∶6∶1的网络结构构建人工神经网络的BP算法模型,相关系数达到0.988.结果表明,与多元线性回归模型相比,BP人工神经网络模型的相关性和预测能力均有较大的提高.
The molecular structure of insect phenoloxidase (PO) inhibitor was characterized by electrical topological state index (En), and multiple linear regression models of 57 PO inhibitor inhibitory activities (pIC50) were established by regression of optimal variables. The non-cross correlation coefficient and cross correlation coefficient were 0.920 and 0.908, respectively, which showed good stability and predictive ability with Jackknife and VIF.The results showed that the main factors affecting the inhibitory activity of PO inhibitor were -OH, O-and C = O molecular structure fragment.Using the three parameters E13, E14 and E16 as the input layer of artificial neural network in the model, set up a 3: 6: 1 network structure to build the BP algorithm model of artificial neural network, The coefficient reaches 0.988.The results show that compared with multivariate linear regression model, the BP neural network model has a greater correlation and predictive ability.