Why are binding-site models more complicated than molecules?

Abstract

A commonly occurring problem in drug development is that the binding affinities for a few compounds to a particular binding site on some protein have been measured, but the crystal structure for that protein is not available. Quantitative structure-activity methods attempt to empirically correlate the binding data with various features of the chemical structures of the drug molecules, so that one can predict the binding of novel compounds and thus aid the search for improved drugs. A common feature of nearly all these methods, however, is that they rely—implicitly or explicitly—on a guess as to the positioning of each molecule when bound to the common site. If one instead assumes that each molecule is free to seek out its optimal positioning in the site, then correlating the observed activity to molecular structure becomes more difficult, and can lead to surprisingly complicated site models. Here we show with some extremely simple artificial examples how this complexity necessarily arises.