Atomic Descriptors And Its Applications To Enantioselectivity Prediction And Atomic Charge Prediction

The topological descriptors were mostly used to describe molecules. However, in this thesis atomic topological descriptors were suggested, and applied to prediction of partial atomic charges and prediction of absolute configuration of major product in asymmetric catalytic reaction, respectively. 1、Studies on prediction of partial atomic chargePartial atomic charge is a simplified representation of electronic distribution between atoms and is closely related to many experimental data, such as NMR chemical shift. However, partial atomic charges can’t be obtained by experiments, and only can be obtained by calculations, e.g., quantum chemistry calculations. Among them, Natural Bond Orbital (NBO) atomic charges are widely accepted and applied to vary of areas. Although NBO scheme based on Density Functional Theory (DFT) can produce high-quality atomic charge, it is high time cost. In this case, quantitative structure-property (QSPR) models were developed to rapidly predict partial atomic charges in this thesis. The process of constructing models are as follows:1)The compounds from ZINC database were partitioned into training set(12541 hydrogen atoms and 22764 non-hydrogen atoms)and test set (4178 hydrogen atoms and 7587 non-hydrogen atoms); 2) NBO charges of all atoms were calculated; 3) 2-Dimensional topological indexes were derived from connections between atoms; 4) Random forest, support vector machine, multilinear regressions and artificial neural networks were used to build models between topological indexes and atomic charges; 5) the atomic charges of test set were predicted by constructed models. For test set, Q2 of hydrogen atoms and non-hydrogen atoms reached to 0.987 and 0.996, respectively. The results indicate that QSPR methods can provide fast access to accurate estimation of NBO atomic charges. 2、Studies on prediction of enantioselectivity in asymmetric reactionsChiral substituent codes are derived from the properties of substituents directly bonded to chiral center. For chiral molecules with one chiral center, chiral substituent code is a molecular descriptor and is also an atomic descriptor of chiral center. In this thesis, the chiral substituent codes would be applied to two data sets:1) 50 pairs of chiral secondary alcohols as reduction products of prochiral ketones using (-)-diisopinocampheylchloroborane as a catalyst; 2) 86 pairs of enantiomers of primary alcohols involved in racemic resolutions by transesterifications or hydrolyses catalyzed by Pseudomonas cepacia lipase. Fisher linear discriminant analysis was used to construct the relationship between chiral substituent codes and enantioselectivities for the two data sets, individually, and the satisfactory results were obtained for both of them. Furthermore, several empirical rules were extracted from the constructed mathematical models, which can almost entirely replace the analytical expressions. The idea has the potential to be generally applied with enantioselective reactions by using different codes and different machine learning methods.