An Atom Counting QSPR Protocol

A simple descriptor, viz the number of carbon (N C )/non-hydrogenic (N NH ) atoms present in a molecule, is proposed for the development of useful Quantitative Structure-Property-Relationship (QSPR) models. This work is prompted by that of Randic and Basak (J. Chem. Inf. Comput. Sci.2001, 41, 614-618) who have shown that this descriptor is superior to the simple connectivity index and is marginally worse than the variable connectivity index for analyzing the toxicity of aliphatic ethers. It is tested in the present work in models pertaining to the estimation of boiling point of alcohols, enthalpy of vaporization of Polychlorinated Biphenyls (PCBs), n-octanol/water partition coefficient of PCBs and chloroanisoles, pK a values of carboxylic acids, phenols, alcohols, etc. High values of various regression coefficients (R 2 , R 2 CV , R 2 AD ) suggest the significance of this descriptor which further improves in the resulting two-parameter QSPR models with electrophilicity or its local variant as an additional descriptor. Other statistical quantities are also analyzed for gaining additional insights into the proposed models. Although the atom number is not the best possible descriptor in a QSPR model in all cases, its simplicity provides it a special place in the development of those models. The present analysis extends the work of Randic and Basak by delineating the situations when N C /N NH performs well vis-a-vis the other descriptors.