Studies On Structure-activity/property Relationship Of Chiral Compounds And Classification Of Protein Structures Prediction

In this thesis, we have mainly investigated the structure-activity/property relationships of chiralcompounds and the prediction of the classification of protein structures. The paper consists of five chapters,and the details of each chapter are as follows.In chapter1, we briefly introduce the importance of the structure-activity relationships of chiralcompounds, several algorithms of QSAR and present research situation of the classification prediction ofprotein structures.In chapter2, the conformation-independent chirality code was implemented by the sum of pi andsigma residual electronegativity as atomic properties in this paper. The chirality codes were applied to adata set of48enantiomeric pairs of chiral amino alcohols that enantioselectively catalyze the addition ofdiethylzinc to benzaldehyde. The relationship models between chirality code of amino alcohol and theabsolute configuration of primary product were constructed by counter-propagation neural network, and theprediction results were satisfactory. For independent test sets,90.0%of catalysts were correctly predictedand89.5%were correctly recognized for training set.In chapter3, the conformation-dependent chirality code was implemented by the sum of pi andsigma residual electronegativity as atomic properties and applied to a dataset of80chiral secondaryalcohols esterified with (R)-MTPA and the corresponding1H NMR chemical shifts. Counterpropagationneural network fed by the conformation-dependent chirality code as the input were trained and the trainedneural network was used to predict the properties of chemical shifts. For the training sets of60chiralsecondary alcohols,96.7%correct recognitions were obtained and the entire test set of20objects were correctly predicted, and91.3%of all objects were correctly predicted for cross-validation of the wholedataset.In chapter4, the studies on absolute configuration of the major product catalyzedenantioselectively by a specific catalyst are one of the important issues of asymmetric reaction. In thischapter, several chirality codes of the secondary alcohols were suggested based on the structure of the twoligands and the properties of the atoms bonded directly to the chiral center. The approached descriptorswere applied to a data set of50enantiomeric pairs of chiral secondary alcohols which were the products ofenantioselective reduction of ketones by DIP-chloride as the catalyst. The relationship models betweenchiral descriptor of secondary alcohol and the absolute configuration of the product were constructed bycounter-propagation neural network. The correct predictions of both training set and test set were not lessthan90%and the cross-validation of leave one-pair out also reached to90%.In chapter5, the information of protein structures and the information of structural classificationof proteins were extracted from the PDB database and from the SCOP database, individually. Then twomethods were suggested to code the protein based on information of three-dimensional coordinate ofproein and the frequences of pairs of amino acids. Finally, the relationship models between the proteincode and the SCOP were constructed by Random Forests, and the results were satisfactory.