Quantitative Structure-Property Relationship (QSPR) models for boiling points, specific gravities, and refraction indices of hydrocarbons

The relationship between physical properties (normal boiling point, specific gravity, and refractive index) of hydrocarbons in the distillate boiling range and their molecular structures was examined using the CODESSA (Comprehensive Descriptors for Structural and Statistical Analysis) program. Multiple linear regression equations with up to eight parameters were used to develop predictive multilinear regression models. Six models for boiling point, specific gravity, and refractive index were obtained for the saturate and aromatic compounds separately. The correlation coefficients (R²) for all six models were >0.99, except for the model of specific gravity for aromatics, which had a correlation coefficient of R² = 0.9881. The standard deviations over the 186-point saturate training set were 6.10 K, 0.007, and 0.004 for boiling point, specific gravity, and refractive index, respectively. Those over the 200-point aromatic training set were 6.30 K, 0.008, and 0.005, respectively. Leave-one-out cross-validation (CV) checks that were performed for all the obtained models led to only slightly smaller correlation coefficients (R² - R_CV² < 0.002). Another set of 34 hydrocarbon compounds was chosen as the validation set to test the saturate models. Aromatic validation sets were also chosen to test the accuracy of the models for boiling point (61 data points), specific gravity (36 data points), and refractive index (27 data points). All the models showed excellent performance, with average errors of <1%. A separate set of three models was obtained for the combined saturates and aromatics data set. Compared to the models obtained for saturates and aromatics separately, these three more-general models were less accurate.