Development of quantitative structure-property relationships using calculated descriptors for the prediction of the physicochemical properties (nD, p, bp, ε, η) of a series of organic solvents

Quantitative structure-property relationship (QSPR) models were derived for predicting boiling point (at 760 mmHg), density (at 25 degrees C), viscosity (at 25 degrees C), static dielectric constant (at 25 degrees C), and refractive index (at 20 degrees C) of a series of pure organic solvents of structural formula X-CH2CH2-Y. A very large number of calculated molecular descriptors were derived by quantum chemical methods, molecular topology, and molecular geometry by using the CODESSA software package. A comparative analysis of the multiple linear regression techniques (heuristic and best multilinear regression) implemented in CODESSA, with the multivariate PLS/GOLPE method, has been carried out. The performance of the different regression models has been evaluated by the standard deviation of prediction errors, calculated for the compounds of both the training set (internal validation) and the test set (external validation). Satisfactory QSPR models, from both predictive and interpretative point of views, have been obtained for all the studied properties.