Quantitative Structure-Activity Relationship Studies On Antitumor Drugs

Cancer is a kind of commonly and frequently encountered disease which severely threaten people's health and life. With the development of science and technology, the occurence of cancer is still in an upward trend. Nowadays, developing anticancer drugs has been paid much more attention by governments of different countries. As an important computational method and common technique in drug design, quantitative structure-activity relationship (QSAR) plays an important role in drug design and development. In recent half century, great impetus has been made by QSAR to the development of synthetic chemistry, pharmaceutical chemistry and drug design, it is proved to be a powerful tool for correlating molecular structure with their physical/chemical property or biological activity.In this thesis, molecular electronegative distance vector (MEDV) and 3D holographic vector of atomic interaction field (3D-HoVAIF) are applied and extended, the former is based on the 2D information of molecular structure, the latter is based on the 3D information of molecular structure. In the modeling process, stepwise multiple regression (SMR), genetic algorithm (GA), multiple linear regression (MLR), partial least squares (PLS) regression, backpropagation artificial neural network (BPANN), support vector machine (SVM) are used to correlate the 2D/3D vector of molecules with their antitumor data, most obtained models have comparable or superior quality compared with literatures. The main contents are as follows:(1) 3D-HoVAIF is used to characterize the molecular structure of benzo[a]phenazines and correlate with their antitumor activity. It is showed from the result that electrostatic, steric and hydrophobic interaction all have contribution to biological activity, correlation coefficient R~2 and R~2CV obtained by cross-validation are 0.854 and 0.601 respectively. Four new compounds are designed and predicted with high activity. A 12-variable QSAR model is constructed for diarysulfonylureas data set, with only 4 principal components (PC) accounting for 92.4% variance of Y data matrices and 76.3% by cross-validation. Besides, a modest modification of numbers and substituted positions of oxygen and sulfur atoms with sp2 hybridization would improve the activity.(2) A 5 variables PLS model is obtained for quinolones data set, which uses 4 PCs to explain 96.0% variance of Y(83.9%, 8.1%, 2.8%, 1.2% for 1st, 2nd, 3rd and 4th PC)...