QSPR/QSAR Studies With The Descriptors Derived From Electrostatic Potentials On Molecular Surface

Quantitative Structure-Activity Relationships(QSAR) plays an important role in drug design especially when the structure of the target is unkown. The parameters of the earlier QSAR models involves the measurement of experiment. With the advent of high throughput screening and combinatorial chemistry, it is possible to synthesize large quantities of compounds in a rather short time. So there is a strong need to be able to predict the property/activity of compounds in that the experimental methods would be very costly and time-consuming. Quantitative Structure-Property Relationships(QSPR) applies the same principle of QSAR to study the relationship between molecular structures and its properties. The QSPR methods were widely used in the fields such as parmaceutical science, materials science, environmental science, etc.Firstly, a profile of QSAR/QSPR is made in the first chapter including the history, the work flow, modeling approaches, structrue parameters and so on. After that, the molecular surface electrostatic potential parameters and their appliances are introduced. Finally, the research ideas of this dissertation are put forward.In the second chapter, QSPR models are established for hydrogen basicity scale pKBHx and the molecular surface electrostatic potential parameters. At first, QSPR models for the pKBHx are established by the means of multiple linear regression(MLR). The hydrogen bond donors(HBDs) are divided into several goups according to the different Hydrogen bond acceptor sites. Linear relationship between hydrogen bond basicity pKBHX of each group and theoretical descriptor Vmin (the spatial mania of molecular electrostatic potential) shows good correlation. Compared with other groups, the nitrogen group QSPR model shows the lowest coefficient of correlation which would improve greatly after elimilating several kinds of HBDs in the nitrogen group. Finally a general QSPR model is built by the means of multiple linear regression(MLR). The stability and predictive power of the general model are validated by rigorous Monte Carlo cross-validation methods.The third chapter focus on the study of hydrophobic substituent/fragment constants using descriptors derived from local electrostatic potentials. Firstly QSPR models are established for the hydrophobic substituent constant(πx) using the theoretical descriptors derived from local electrostatic potentials at the substituent atoms. All of the six descriptors employed in the QSPR models have definite physical meanings including Vmin, Vs,max,ΣVind,+, Nindtot,Πind and σind2,-. These descriptors are found to be related to hydrogen-bond basicity, hydrogen-bond acidity, cavity, or dipolarity/polarizability terms in linear solvation energy relationship(LSER), which endows the models good interpretability. A rigorous Monte Carlo cross-validation is performed to confirm stabilities and predictive powers of the constructed models. Then, eight groups of meta- or para- disubstituted benzenes and one group of 3- or 4-substituted pyridines are investigated. QSPR models for individual systems are achieved with a combination of above six descriptors. The statistical qualities of the models are generally inferior to those for benzene system. Finally, two QSPR models are established for Rekker’s fragment constants (foct).It has been demonstrated that the descriptors derived from electrostatic potentials at the fragments, can be well used to quantitatively express the relationship between fragment structures and their hydrophobic properties, regardless of the attached parent structure or the valence state.In the fourth chapter, four linear QSPR models are establised bewteen the four parameters(i.e. S, A, B, E) of linear solvation-energy relationship and electrostatic potential descriptors. It turns out that the MLR models for S and A both show excellent stabilities and predictivity powers. Moreover the MLR models for B and E are good as to the stabilities and predictivity powers. Then, the gaussian process(GP) models and least square support vector machine(LSSVM) models for the four parameters are constructed. As to the parameters A and S, the linear models are better than GP models and LSSVM models. However, the situations become different for the parameters B and E. Compared with the MLR models, the GP models for B and E show better stabilities and predictivity powers than the MLR models.The fifth chapter is devoted to establishing QSAR models between monoamine (i.e.5-hydrotryptamine, dopamine, noradrenergic) transporter inhibitor activities and surface elecrostatic potential descriptors. It turns out that the MLR and GP models for 5-hydrotryptamine transporter inhibitors both show pretty good stabilities and predictivity powers and moreover, the GP model is slightly better than the MLR model. However the linear and nonlinear models for dopamine and noradrenergic show fair predictivities. The qualities of linear and nonlinear models for dopamine and noradrenergic are generally inferior to the qualities of MLR and GP models for 5-hydrotryptamine.