A General Method for Exploiting QSAR Models in Lead Optimization

Abstract

Computer-aided drug design tools can generate many useful and powerful models that explain structure-activity relationship (SAR) observations in a quantitative manner. These models can use many different descriptors, functional forms, and methods from simple linear equations through to multilayer neural nets. Using a model, a medicinal chemist can compute an activity, given a structure, but it is much harder to work out what changes are needed to make a structure more active. The impact of a model on the design process would be greatly enhanced if the model were more interpretable to the bench chemist. This paper describes a new protocol for performing automated iterative quantitative structure-activity relationship (QSAR) studies and presents the results of experiments on two QSAR sets from the literature. The fundamental goal of this work is to try to assist the chemist in his search for what to make next.