The Structure Of Small Molecules And The Melting Point And Protein Affinity Quantitative Structure-activity Relationship Studies,

To meet the scientists' increasing needs of chemical knowledge from large-scale data sets, Chemoinformatics comes in.This paper was organized as following:The first chapter indroduced the main databases which included the Chemical and the Biological databases.In the second chapter, two datasets were compared with each other to illustrate a basic problem appeared in the research field of QSAR/QSPR often. A new method of local lazy regression (LLR, which obtains a prediction for a query molecule using its local neighborhood, rather than considering the whole data set) was used to try to improve the effect of prediction. The result showed that LLR could express the local features exhibited by some groups of molecules and describes the local relationships between their structure and properties or biological effects completely. And the result from LLR is much better than the one from global model.In the chapter 3, CoMFA (Comparative Molecule Field Analysis) was used to build the 3D-QSPR model for melting point. Both of the electrostatic potential field and van der Waals field could explain the interaction among molecules in crystal and describe the property of melting point of compounds completely. They were supplements for each other. It showed that melting point 3D-QSPR analysis can benefit from the method of CoMFA.A newly developed descriptors of three-dimensional holographic vector of atomic interaction field (3D-HoVAIF) which was on the base of molecular electronegativity-distance vector (MEDV) were employed to study the three dimensional quantitative structure-activity relationships (QSAR) for binding affinity of 41 flavonoids at Benzodiazepine site in GABA_A receptors, meanwhile the descriptors of MEDV were utilized to generate two dimensional QSAR model for the biological activity of these flavonoids. And then, other 3D-QSAR model was generated by the method of CoMFA. The results of the models showed that the newly developed descriptors could express structural information related to biological activity of flavonoids successfully, and the 2D model was better than the 3D model.In the chapter 4, two structural similarities clustering method were used to screen a large drug database. One was a statistical method based on same fragments between each pair of compounds and it adopted Tanimoto formula to measure structural similarity, the other was maximal common sub-structural (MCS) similarity clustering. PCR and PLS Monte Carlo cross validation were taken to build models for subsets which were screened out from the large database. The correlation coefficients of training sets for all of the subsets were 0.9 nearly and the correlation coefficients of test sets for all of the subsets reached 0.8. The forecast average absolute error was only 25K. The correlation coefficients of both training set and test set for the database were 0.65 and the forecast average absolute error reached 40K nearly in literature where using artificial neural network methods.In the fifth chapter, a new concept of structural similarity which was based on the similarity of properties between the compounds was put forward. It pointed out that the the structural similarity was different for the homologous compounds with different properties.