Abstract
Two general problems of QSAR studies are discussed. The first is related to the fact that, in general, the studied property is an integral result of multistage interactions carried out at various solvent environments with different polarity under which different conformers should be prevalent. In contrast with the conventional QSAR methodologies which provide a single 3-D molecular model for each 2-D structure in the modeling process, the newly introduced dynamic QSAR method is aiming to mimic the multiplicity of 3-D molecular shapes taken from the chemicals at the different stages of the processes conditioning the endpoint under investigation. The second QSAR problem concerns the proliferation of molecular descriptors used for QSAR studies. Hoping to account for any structural factors that may affect the property under investigation, the QSAR studies often generate a large set of topological, steric, and electronic parameters. This problem can be handled by two different approaches. One can incorporate purely statistical methods, such as the multiple regression analysis, principle component analysis (PCA), partial least square (PLS), etc. In addition to the combinatorial complexity in model derivation, the large dimensionality of the initial data matrix can increase the risk for a chance correlation. The drawback of PCA, PLS and related methods is the loss of physical meaning of the resulted orthogonal regressors. We turn user's attention on QSAR methods based on physico-chemical principles used for parameter screening. In fact, one selects potentially significant factors for the studied property by creating hypotheses on interaction mechanism. The larger the knowledge for interaction mechanism the smaller the number of hypotheses created.
Key Words: