【作者】 杨善彬；

【导师】 夏之宁； 李志良；

【作者基本信息】 重庆大学 ， 生物医学工程， 2008， 博士

【摘要】 本文结合传统分子结构表征（MSC）技术的优点提出了一种全新的3D结构描述子:三维原子场全息作用矢量（ Three Dimensional Holographic Vector of Atomic Interaction Field, 3D-HoVAIF）。该方法从分子立体结构的两个空间不变量——原子相对距离和原子自身性质出发,基于三种常见的与生物活性直接相关的原子间非键作用方式（静电、立体和疏水作用）,在不借助任何实验参数及无需样本构象重叠的前提下得到了用于表征药物分子结构特征的三维矢量描述子。使用Chemoffice 8.0构建待研究样本的分子立体结构,并使用Chem3D自带的MOPAC半经验量子化学软件在AM1水平上优化得到分子结构,并采用Mulliken布局分析法以单点（single-point）形式计算出原子的净电荷数量,将分子中每个原子的空间位置及电荷分别以笛卡儿坐标和净电子数目的形式输入到由C语言编写的应用程序中加以处理得到3D-HoVAIF描述子。三维原子场全息作用矢量最多含有165个描述子,由于缺少一些原子类型,所以对于大多数体系,作用矢量少于165个。然后运用逐步线性回归（SMR）筛选变量后,用多元线性回归（MLR）,偏最小二乘（PLS）等统计工具建立药物定量结构-活性关系（QSAR）模型,取得了较满意的结果。该方法创新之处在于无须任何实验参数及无需样本构象叠合、计算简便迅速、物理化学意义明确并且避免了当今大多3D药物设计技术在应用前需要对分子进行叠合操作等方面缺点;另外3D-HoVAIF按族对原子进行分类使其可以适用于含有多种杂原子的复杂分子体系。（1）采用3D-HoVAIF对19个喹诺酮类化合物进行结构表征。采用逐步回归对变量进行筛选后,运用偏最小二乘回归建立了3D-HoVAIF描述子与喹诺酮类化合物抗肿瘤活性之间的QSAR模型,该模型的复相关系数（R²）、交互校验复相关系数（Q²）和模型的拟和均方根误差（RMSEE）分别为R²= 0.912, Q_cum²= 0.811,RMSEE= 0.284。表明三维原子场全息作用矢量能有效地提取该类分子结构信息并与其生物活性呈良好线性关系。（2）采用3D-HoVAIF对26个噻唑并[2,3-d]嘧啶-4-酮类化合物的6种抗菌活性进行了系统的QSAR研究,建立了SMR-PLS定量构效关系模型,所建模型的复相关系数（R²）、交互校验复相关系数（Q²）和模型的拟和均方根误差（RMSEE）分别为R²≥0.891, Q_cum²≥0.85和RMSEE≤0.174,得到的6个模型预测能力和稳定性均较好,优于文献利用拓扑指数建立的Hansch分析结果。（3）采用3D-HoVAIF对68个甲状腺激素受体（TR）配体化合物的化学结构与活性建立定量构效关系。对化合物进行了结构参数化表达, 55个样本为训练集,13个样本为预测集,采用SMR-PLS建立了定量构效关系模型,模型的复相关系数和交互检验复相关系数和拟和均方根误差分别为R²=0.766, Q_cum²=0.586,RMS=0.641 （TRα）; R²=0.754, Qcum2=0.593,RMS=0.672 （TRβ）,与文献的二维结果q²=0.781 （TRα）和q²=0.693 （TRβ）相当,但本模型物理意义明确。（4）采用3D-HoVAIF对123个流感病毒神经氨酸酶抑制剂建立定量构效关系,运用逐步线性回归（SMR）筛选变量后,用多元线性回归（MLR）建模的得到了9变量模型,相关系数为R=0.885,留一法检验的复相关系数为RCV2=0.736。显然,3D-HoVAIF能够较好地表征神经氨酸酶抑制剂的构效关系,所建模型具有良好的内部估计能力和外部预测能力。为了深入分析3D-HoVAIF对神经氨酸酶抑制剂样本集的表达和建模性能,采用D-优化算法将123个抑制剂分为两部分,即训练集和测试集各为100和23个样本。同样用3D-HoVAIF对100个神经氨酸酶抑制剂进行结构表征,然后采用逐步回归对变量进行筛选后,运用偏最小二乘建立3D-HoVAIF描述子与神经氨酸酶抑制剂活性之间的QSAR模型。结果:复相关系数（R²）,交互校验的复相关系数（Q2）和模型的标准偏差分别为R²=0.805,Q2=0.657,SD=0.936,模型具有良好的稳定性和预测能力,并对文献中23个药物和设计的32个化合物进行了预测。表明三维全息原子场作用矢量能较好表征该类分子结构信息,值得进一步推广应用。（5）基于分子三维结构信息的三维原子场全息作用矢量（3D-HoVAIF）的构效关系研究,再根据流感病毒神经氨酸酶（NA）的活性点以及与神经氨酸酶抑制剂（NAI）的缀合特性并参考已有研究成果,设计出了11个系列共计82个化合物,其中有的预测活性较已上市的奥司米韦（oseltamivir, Tamiflu?）要高,这些设计化合物母环结构分别是环己烯系、呋喃系、吡唑系和吡嗪系,母环为吡啶和嘧啶系的设计化合物的预测生物活性与奥司米韦接近,它们均将作为重点后续研究对象。更多还原

【Abstract】 A novel molecular structural characterization （MSC） method, three dimensional holographic vector of atomic interaction field （3D-HoVAIF）, is proposed in our laboratory, focusing its idea on three points: （a） Atoms are typed for 10 kinds according to their families in periodic table of elements and self hybridization states; （b） Three non-bonded （electrostatic, van der Waals and hydrophobic） factors, directly related to bioactivities, are utilized to express intramolecular potential energies; （c） Based on molecular steric structures, 165 non-bond interaction items calculated are taken as the three-dimensional （3D） structural descriptors of the molecule. Original spatial structures of the target compounds are autogenerated by software Chemoffice 8.0, then implementing a molecular mechanic （MM） conformation optimization （adopting MM+force field） with molecular simulation software HyperChem 7.5, and the end semi-experimental quantum chemistry software MOPAC 6.0 is further utilized to generate the ultimate 3D structures at AM1 levels. Simultaneously, atomic partial charges are calculated by Mülliken population analysis. Taking forms of Cartesian coordinates and partial charges respectively, spatial position for each atom in a molecule and the atomic charges are input into C-edited program Super-3D.EXE, giving rise to 3D-HoVAIF descriptors of the molecules. In the modeling process, stepwise multiple linear regression （SMR）, multiple linear regression （MLR）, partial least squares （PLS） regression, are used to correlate the 3D vector of molecules with their dissolved data, most obtained models have superior quality compared with literatures.（1） The developed three-dimensional holographic vector of atomic interaction field （3D-HoVAIF） was used to describe the chemical structures of 19 antitumor quinolone agents. After the structural characterization, the descriptors obtained were screened by stepwise multiple regression （SMR） then a model of 3D-HoVAIF descriptors and antitumor quinolone agents’activity was built with partial least square regression （ PLS）. The obtained model with the cumulative multiple correlation coefficient （R²） , cumulative cross-validated （Q²） and the Root Mean Square Error of Estimation （RMSEE） were R²=0.912, Q_cum²=0.811, RMSEE=0.284 respectively. The result confirmed that 3D-HoVAIF is able to extract molecular steric potential information efficiently,and well to relate with the bioactivities.（2） And the developed descriptor, three dimensional holographic vector of atomic inter-action field（3D-HoVAIF）, was also used to describe the chemical structures of 26 thienopyrimidones. After variable screening by stepwise multiple regression （SMR） technique, a partial least square （PLS） regression model was built with 3D-HoVAIF. The model is satisfactory comparing to reference since correlation coefficients of molecular modeling（R²）, cross-validation(Q_cum²) and root-mean-square error of estimations（RMSEE） are≥0.891,≥0.85 and≤0.174, respectively, showing that the model has favorable estimation and prediction capabilities.（3） To study the relationships between the chemical structures of Thyroid receptors ligands with activities, the developed three-dimensional holographic vector of atomic interaction field （3D-HoVAIF） was used, 55 samples for the training set and 13 samples for the test set. After the structural characterization, the descriptors obtained were screened by stepwise multiple regression （SMR）, correlation coefficients and crossvalidated correlation coefficients were obtained , R²=0.766,Q2=0.586, RMS=0.641 （TRα） and R²=0.754, Q²=0.593, RMS=0.672 （TRβ） respectively. These results show that the model have favorable stability and good prediction capability and the 3D-HoVAIF is applicable to the molecular structural characterization and biological activity prediction, suggesting that the models could be useful in the design of novel, more potent TR ligands.（4） To study the quantitative structure-activity relationship QSAR of 100 influenza neuraminidase inhibitors, three-dimensional holographicvector of atomic interaction field （3D-HoVAIF） method was used to describe the chemical structure of influenza neuraminidase inhibitors. After the structural characterization, the descriptors obtained were screened by least square regression （ PLS）. The obtained model with the cumulative multiple correlation coefficient (R_cum² ) , cumulative cross-validated （Qcum2 ） , and standard error of estimation （ SD） were R_cum² = 0.805, Q_cum² = 0.657 and SD = 0.936, respectively. The result shows that the model has favorable stability and good prediction capability and predicted the literature 23 drugs as well as design of 32 compounds. The 3D-HoVAIF is applicable to the molecular structural characterization and biologicalactivity prediction.（5） Finally, using 3D-HoVAIF method, 11 series of the designed neuraminidase inhibitors were predicted , according to the active sites of influenza neuraminidase. Some of them may be higher activities than oseltamivir, an approved neuraminidase inhibitor recently. This designed compounds will be studied in the future in our laboratory.更多还原