OBJECTIVE MODELS FOR STEROID-BINDING SITES OF HUMAN GLOBULINS

We report the application of a recently developed alignment-free 3D QS AR method [Crippen, G.M., J. Comput. Chem., 16 (1995) 486] to a benchm ark-type problem. The test system involves the binding of 31 steroid c ompounds to two kinds of human carrier protein. The method used not on ly allows for arbitrary binding modes, but also avoids the problems of traditional least-squares techniques with regard to the implicit negl ect of informative outlying data points. It is seen that models of con siderable predictive power can be obtained even with a very vague bind ing site description. Underlining a systematic, but usually ignored, p roblem of the QSAR approach, there is not one unique type of model but , rather, an entire manifold of distinctly different models that are a ll compatible with the experimental information. For a given model, th ere is also a considerable variation in the found binding modes, illus trating the problems that are inherent in the need for 'correct' molec ular alignment in conventional 3D QSAR methods.