Self-organizing neural networks for modeling 3D QSAR--a comparative study.

Abstract

Different architectures of self-organizing neural networks (SOM) have been used for modeling 3D QSAR. The atomic coordinates and partial atomic charges were used as input signals. In particular, the steroids complexing corticosteroid binding globulin (CBG) that are used as a benchmark measuring the performance of drug design methods have been applied to compare between individual methods. The sensitivity of the different architectures for changes of the alignments of the molecules within series, as well as the possibility for alignments based on the molecular inertial axes have been tested.