Development of validated quantitative structure-retention relationship models for retention indices of plant essential oils.

Quantitative structure-retention relationship (QSRR) models were developed for the retention indices of 505 frequently reported components of plant essential oils. Multiple linear regression was used to build QSRR models for the dimethyl silicone, dimethyl silicone with 5% phenyl groups, and polyethylene glycol stationary phases. We tried to improve the variable selection and modeling method based on prediction method for selecting the optimum descriptors from the molecular weight, 75 topological indices, and 170 atom-type E-state indices. The three-variable QSRR models perform high correlation coefficients of 0.937 for dimethyl silicone and 0.933 for dimethyl silicone with 5% phenyl groups stationary phase. Four variables were selected to developed QSRR model for the polyethylene glycol stationary phase. The leave-one-out and leave-many-out cross-validations, bootstrapping, and y-randomization test showed the three models are robust and have no chance correlation. The external validation with the test set showed the three models present high externally predictive power. The three models presented high-quality fit, internally, and externally predictive power. It is expected that the models can effectively predict retention indices of essential oils components without experimental value.