| 冯长君,杨杰元,杨雪颖,杨沛艳,冯惠.化学通报,2022,85(10):1249-1254. |
| 基于电性距离矢量研究多氯联苯生物降解速率常数 |
| Study on the biodegradation rate constants of polychlorinated biphenyl compounds based on electronegativity distance vector |
| 投稿时间：2022-01-03 修订日期：2022-03-31 |
| DOI： |
| 中文关键词: 多氯联苯 生物降解速率常数 电性距离矢量 人工神经网络 定量结构-生物降解性关系 |
| 英文关键词:polychlorinated biphcnyl rate constant for biodegradation electronegativity distance vector artificial neural network back propagation quantitative structure-biodegradability relationship(QSBR) |
| 基金项目:国家自然科学基金项目(21075138)、结构化学国家重点实验室开放基金项目(2016003)、江苏省大学生创新创业训练项目(xcx2020143)和徐州工程学院大学生创新创业训练项目(2019013)资助 |
|
| 摘要点击次数: 719 |
| 全文下载次数: 0 |
| 中文摘要: |
| 通过多元线性回归和人工神经网络方法建立66种多氯联苯生物降解速率常数(K1)的定量构效关系(QSAR). 基于电性距离矢量(Mk)，建立了lnK1的最佳三参数(M91、M25和M15)线性模型，其传统相关系数(R2)、交叉验证系数(Rcv2)分别为0.833、0.809。经R2、Rcv2、VIF、FIT、AIC检验，所建模型具有较强的稳定性和良好的预测能力. 将M91、M25、M15作为人工神经网络的输入层结点，采用3:10:1的网络结构，利用BP算法获得了一个令人满意的lnK1模型，训练集、验证集、测试集和总体的R2依次为0.991、0.995、0.997和 0.993。与多元线性回归模型相比，非线性lnK1-BP模型具有更好的预测能力。这两种回归方法相辅相成，线性回归方法为神经网络模型提供了具体的物理解释，而神经网络方法为线性模型提供了更准确的预测结果。 |
| 英文摘要: |
| The quantitative structure-activity relationship (QSAR) models was established by using linear regression method (multiple linear regression) and nonlinear regression method (artificial neural network back propagation) to estimate and predict the biodegradation rate constants(K1) of 66 polychlorinated biphenyl compounds(PCBs). On the basis of electronegativity distance vector(Mk), the best three-parameter (M91, M25 and M15) QSAR model of lnK1 is constructed. Traditional correlation coefficient (R2) and cross-validation correlation coefficient (Rcv2) of the model are 0.833 and 0.809, respectively. According to R2, Rcv2, VIF, FIT, AIC and F tests, the model has high reliability and good prediction ability. It can be seen from the three parameters of the model that the main factors affecting the biodegradation rate constant are molecular structure fragments: =CH– (secondary carbon), =C< (tertiary carbon) and chlorine atom (Cl–) in aromatic ring. It shows that the main factor affecting K1 is the number of chlorine atoms in aromatic ring, followed by the position of chlorine atoms in aromatic ring. The topological descriptors (M91, M25 and M15) are the input nodes of artificial neural network, and the biodegradation rate constant (lnK1) is the output node, forming a 3:10:1 network structure, and using BP algorithm to obtain a satisfactory lnK1-BP model. The correlation coefficients (R2) of training set, test set, verification set and population are 0.991, 0.995, 0.997 and 0.993, respectively. Compared with multivariate linear regression model, nonlinear lnK1-BP model has better prediction ability. The two regression methods complement each other. The linear regression method provides a specific physical explanation for the neural network model, while the neural network method provides a more accurate prediction result for the linear model. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|