【作者】 戴益民；

【导师】 刘又年；

【作者基本信息】 中南大学 ， 应用化学， 2012， 博士

【摘要】 化合物结构-性质／活性定量相关(定量构效关系,Quantitative Structure-property/activity Relationship, QSPR/QSAR)研究,最初是作为生物领域的一个研究分支,是为了适应合理设计生物活性分子的需要而发展起来的。目前它已成为分子设计与目标化合物研发的基础课题和重要环节,也是对化学品进行环境毒性评价的重要方法。它主要应用各种统计学方法和分子结构参数研究化合物的结构与其各种物理化学性质以及生物活性之间的定量关系。本论文从分子设计角度出发,运用原子平衡电负性原理结合分子结构参数来定量估计并预测化合物性质、生物活性及环境毒性。具体研究内容如下：1.综述了定量构效关系研究现状、分子设计及分子模拟的基本方法和原理、电负性均衡原理、原子电荷计算方法以及相关方法应用研究的进展。2.基于分子图论提出了一种用于表征咪唑啉衍生物缓蚀剂分子局部化学微环境及原子杂化状态的新颖结构描述子：电性连接性指数0Kv、1Kv和咪唑啉环非氢原子平衡总电荷分数MCI,研究其对15种咪唑啉类缓蚀剂抗CO2、H2S腐蚀性能的定量构效关系。结果表明,模型计算值、留一法交互检验预测值的复相关系数分别为0.9764、0.9546,所建模型具有良好的稳定性和优异的外部预测能力；同多元回归方法比较,运用人工神经网络法其复相关系数为0.9848,相关结果得到较大改善。增加咪唑啉环上取代基长度、减小分子支化度和降低咪唑环非氢原子平衡总电荷分数能显著提高咪唑啉衍生物缓蚀剂的缓蚀性能。3.在距离矩阵的基础上采用原子的平衡电负性和化学键相对键长校正含有多重键的化合物,提出了两个新颖的拓扑电负性指数YC、WC,同时结合路径数P3对92个碳氢化合物的局部化学微环境进行结构表征,并对化合物的闪点进行了QSPR研究。采用多元线性回归得到训练集模型的复相关系数和标准偏差分别为0.9923和5.28,模型实验值与计算值的平均绝对误差仅3.86K,相对误差1.46%。同时采用内部及外部双重验证的办法对所建模型进行检验,留一法(LOO)检验和训练集、检验集闪点的预测值和实验值较为吻合,结果表明模型具有良好的内部稳定性和外部预测能力。4.采用新颖的原子拓扑矢量YC、原子平衡电负性χe、结构信息参数[NHi(i=α、β)]和γ校正参数对63个无环饱和脂肪醇的局部化学微环境进行了结构表征,并对化合物13C NMR化学位移进行了定量结构-波谱关系(QSSR)研究。采用偏最小二乘回归得到模型的复相关系数R和标准偏差S分别为0.9915和2.4827,对353个碳原子13C NMR化学位移的实验值与计算值的平均绝对误差仅为2.01ppm。同时,采用留分法和外检验方法测试模型的内部稳定性和外部预测能力。另外从分子结构出发提出四个分子结构描述符YC、χe、[NHi(i=α、β)],运用多元线性回归方法建立55个醇碳原子13C核磁共振谱的定量结构-波谱关系模型。结果表明,模型复相关系数和标准方差分别为R2=0.9824和S=0.8698。同时采用留一法进行检验,结果表明模型具有良好的稳定性和预测能力,优于目前文献的研究结果。5.将距离矩阵与邻接矩阵相结合提出了新颖的表征多环芳烃分子支化度大小的描述子CN和表征多环芳烃分子结构大小的描述子CT,采用多元线性回归方法构建了100种多环芳烃气相色谱保留指数的定量相关模型。所得模型相关系数R=0.9970,交叉验证相关系数RCv=0.9967。随机选出70种多环芳烃化合物作为训练集,其余作为测试集来验证模型的预测能力和稳健性。结果表明：训练集和测试集的复相关系数分别为0.9972和0.9968,定量计算结果与实验测定值符合较好,优于目前文献的研究结果。6.采用量子化学描述符建立蛋白同化雄性激素类固醇半波还原电位的定量构效关系模型。描述符由半经验方法计算所得,使用偏最小二乘法(PLS)和反向传播神经网络(BP-ANN)成功建立了线性和非线性相关模型。通过定量结构-电化学定量关系(QSER)研究表明：蛋白同化雄性激素类固醇的描述符和半波还原电位存在显著相关性,相关模型的稳定性和预测能力采用留一法交互检验和外部测试法来完成,该研究可成功用于分析鉴定真正意义上的雄性激素类固醇药物。更多还原

【Abstract】 Quantitative structure-property/activity relationship (QSPR/QSAR) was originally introduced as a branch in the biological field and developed in response to rational design of bioactivity molecules. At present, QSPR/QSAR research had become a basis topic and important tache for molecular design and R&D of new goal compounds, and was also an important assessment method of environmental toxicity for chemicals. It had been widely used for the prediction of various physicochemical properties and biological activities of organic compounds by using different statistical methods and various kinds of molecular descriptors. In this thesis, based on the molecular design, atomic equilibrium electronegativity and molecular structrural descriptors were utilized to establish the QSPR/QSAR models in order to estimate and predict compound properties, biological activities and environmental toxicties. The main contents and conclusions were given as follows:1. In this paper, a brief review of principle, research methods and current status for QSPR/QSAR, molecular design and molecular modelling, atomic equilibrium electronegativity and atomic charge were presented. In this section, the research progress of applications in QSPR/QSAR, molecular design and molecular modelling, equilibrium electronegativity and atomic charge were introduced in detail.2. Based on the molecular graphic theory, novel molecular structure descriptors of electrical connectivity index0Kv,1Kv and the imidazoline ring of non-hydrogen atoms balance total charge fraction (MCI) was proposed for expression of local chemical microenvironment and atomic hybridation state. A quantitative structure-property relationship (QSPR) of estimating fifteen imidazoline corrosion inhibitors efficiency (CIE) for anti-corrosion behavior towards hydrogen sulfide and carbon dioxide was established including descriptors0Kv,1Kv and MCI. The results showed that correlation coefficient of modelling calculated and leave-one-out cross-validation (LOO-CV) predicted value were0.9764and0.9546, respectively. The QSPR model was of good stability and external predictive capability. For the same purpose, artificial neural network was applied and the result was improved. The results proposed that increasing substitution length of the imidazoline ring, reducing the molecular branching and lowering the imidazoline ring of non-hydrogen atoms balance total charge fraction had a significant effect.3. Two novel topological electro-negativity indices based on distance matrix, named YC and We indices, were put forward and could be used for modelling properties of multiple bond organic compounds by equilibrium electro-negativity of atom and relative bond length of molecular. A quantitative structural property relationship (QSPR) model for estimating flash point of92compounds was developed based on our newly introduced topological electro-negativity indices Yc and WC and path number parameter P3. The model correlation coefficient and standard error for training set in multiple linear regression were0.9923and5.28, respectively. The average absolute error of flash point was only3.86K between experimental values and calculated values, the relative error was1.46%. Furthermore, the model was strictly analyzed by both internal and external validations. The predicted values were obtained in good agreement with experimental values for leave-one-out (LOO) and the training set and validation set. The results showed that this QSPR model was of good stability and powerful prediction ability.4. A newly developed topological vector of atom Yc, equilibrium electro-negativity of atom Xs, molecular structural information parameter [NiH(i=α、β)] and y calibration parameter were used to describe the local chemical microenvironment of63acyclic alcoholic compounds. A quantitative structural spectrum relationship (QSSR) was systematically studied between13C NMR chemical shifts of353carbon atoms and their molecular structure descriptors. By partial least regression (PLS), the statistical results indicated that the model correlation coefficient and standard error were0.9915and2.4827, respectively. And the average absolute error was only2.01ppm between the calculated and experimental chemical shifts for353carbon atoms. To validate the estimation stability for internal samples and the predictive capability for external samples of resulting models, leave-molecule-out cross validation and external validation were performed. Compared with the reported result, not only the number of descriptors employed in this paper was much fewer, but also the calculation was much easier. In addition, a quantitative structure-spectrum relationship model was developed to simulate13C NMR spectra on carbinol carbon atoms for55alcohols. The proposed model, using multiple linear regression, contained four descriptors Yc, Xe,[NiH(i=α、β)] solely from the molecular structure of compounds. The statistical results of the final model showed that R2=0.9824and S=0.8698. The model was statistically significant and showed very good stability to data variation using the leave-one-out cross-validation. The comparison with the other approaches also revealed good behaviors of our method in this QSSR study.5. Two novel molecular structure descriptors based on distance matrix and adjacency matrix, named CN and CT were proposed which characterized branch vertex and molecular structural size of polycyclic aromatic hydrocarbons (PAHs), respectively. A quantitative structure-retention relationship (QSRR) model for estimating gas chromatography retention indexes of100polycyclic aromatic hydrocarbons was constructed by multiple linear regression (MLR). A satisfactory result was obtained that the correlation coefficients in partial least square and cross validation using leave-one-out were0.9970and0.9967, respectively. In order to verify the prediction ability and stability of the model, the samples were divided into70training set and30test set randomly. The result indicated the correlation coefficients of training set and test set were0.9972and0.9968, respectively. The quantitatively calculated results were in agreement with experimental ones basically. The model was compared with recently proposed QSRR models of the similar data. It was found that the present model was all better than relevant achievements in literatures.6. A quantitative structure-electrochemistry relationship (QSER) study of anabolic androgenic steroids had been done on the half-wave reduction potential (E1/2) using quantum and physicochemical molecular descriptors. The descriptors were calculated by semi-empirical method. Successful models were established using partial least square (PLS) regression and back-propagation artificial neural network (BP-ANN). The QSER study results indicated that the descriptors of these derivatives had significant relationship with half-wave reduction potential. The stability and prediction ability of these models were validated using leave-one-out cross-validation and external test set. This study might be helpful in the future successful identification of "real" or "virtual" anabolic androgenic steroids.更多还原