Estimation of water solubility from atom-type electrotopological state indices

Based on the atom-type electrotopological state (E-state) indices, a quantitative structure-property relationship model for the prediction of aqueous solubility for a diverse set of 745 organic compounds is presented. The multiple linear regression analysis was used to build the models. A training set of 674 compounds, containing 349 liquids and 325 solids and having a range of aqueous solubility (log S) values from 2.77 to 11.62, was obtained from the literature. For this set, the squared correlation coefficient and standard deviation for a linear model with 31 atom-type E-state indices and three simple correction factors were r2 = 0.94 and s = 0.58 (log units), respectively. The corresponding statistics for the test sets not included in the training set were, for a set of 50 pesticides, r2 = 0.79 and s = 0.81 and, for a set of 21 drug and pesticide compounds, r2 = 0.83 and s = 0.84, respectively. The contribution of melting points was also evaluated. The use of melting point increased the accuracy of the models in the fit of the training set but not in the prediction of the test sets. Hence, the proposed method offers fast and accurate estimation of aqueous solubility of organic compounds using atom-type E-state indices without the need of any experimental parameters like the melting points.