Molecular structural index control in property-directed clustering and correlation

Abstract

The correlation of physicochemical and pharmacological properties with molecular structure is often difficult because plots of the data apear to have a random distribution. While this may be true in some cases, investigators may suspect inherent relationships, and indeed, all properties are ultimately dependent on specific molecular structure. This was the case in attempting correlations between Hammett substituent constants and calculated dipole moments and molecular transform and normalized molecular moment structure indices. However, recent correlations of pharmacological and structural indices had shown that a primary numerical sorting of the structure indices with a secondary sorting of the biodata resulted in a natural clustering that rationally defined the data and permitted linear correlations within clusters that contained sufficient points. In the present case, a series of meta- and para-substituted benzoic acids were considered, and it was shown that the Hammett constants were linearly related only to the normalized charge moment index. However, application of the clustering paradigm permitted correlation of other structural indices with both the Hammett constants and dipole moments. These results and the generalized employment of this methodology are discussed. © 2000 John Wiley & Sons, Inc. Int J Quant Chem 80: 1164–1171, 2000