Three Dimensional Molecular Structure-Biological Activity Correlation by Pattern Recognition

A molecular structure transform has been developed that allows the three-dimensional structure of molecules to be related to their biological activity. The transforms utilize principal component analysis to transform the three dimensional cartesian coordinates and any atomic information such as electronegativity for each atom in a molecule. A predetermined and constant number of features is extracted from each molecule which can be analyzed by the methods of pattern recognition in order to relate molecular structure to any molecular function; in this case, carcinogenicity. The transform is graphically demonstrated using a simple triatomic molecule and then applied to two divergent classes of carcinogenic molecules; dimethylaminoazobenzenes and polycyclic aromatic hydrocarbons. Structure-activity results are presented along with an exciting potential for drug design. (Author)