RESOLUTION OF AMBIGUITIES IN STRUCTURE - PROPERTY STUDIES BY USE OF ORTHOGONAL DESCRIPTORS

We initiate here a systematic analysis of molecular descriptors, the quantities used to represent molecules in structure-property and structure-biological activity studies. We illustrate the approach with Hosoya's Z index, which is based on the enumeration of disjoint edges in molecular graphs. The purpose of this analysis is to discern similarties and differences among the molecular descriptors used in regression analysis. In this way, we hope to facilitate the rational use and perhaps the rational design of such descriptors. The main tool in this analysis is the recently described methodology of construction of orthogonalized molecular descriptors in multivariate regression analysis. The analysis points to critical parts of descriptors which are responsible for their performance in a regression.