Prediction of aqueous solubility and the octanol-water partition coefficient for lipophilic organic compounds using molecular descriptors and physicochemical properties

The utility of thirty-nine molecular descriptors and physicochemical properties to model the solubility (S) and octanol-water partition coefficient (Kow) of thirty-one lipophilic organic compounds was assessed using least squares linear regression analysis. The modelling of Kow when all the compounds were treated together or in groups of structurally related compounds was adequate however, utility for prediction was greatly improved by separate treatment of ionizing and non-ionizing compounds. The log Kow values of the ionizing compounds could best be predicted by multi-parametric linear regression (MLR) equations utilising pKa and molecular weight. The linear regression equations for log S were not as significant as those for log Kow. Several novel molecular descriptors that yielded high correlation coefficients in linear regression equations were the approximate sigma electron density, radius of gyration and the first and second principle moments of inertia. Other useful properties were the sum of all absolute valency charges, density at 20°C, liquid density and pKa.