QSPR Models for Predicting Retention Indices of Polygonum minus Huds. Essential Oil Composition Using GA-BWMLR and GA-BPANN Methods

Abstract

In the present study, quantitative structure-property relationship (QSPR) models were used for modeling and predicting retention index (RI) values of chemical composition in Polygonum minus essential oil. In gas chromatography, RIs are used to convert retention times into system-independent constants. The dataset of 40 molecules was divided into training and external validation sets. A set of molecular descriptors were calculated from the optimized structures of the molecules using Gaussian 09 and Dragon software. The genetic algorithm (GA), backward stepwise multiple linear regression (BW-MLR), and back-propagation artificial neural networks (BPANN) were used to obtain suitable QSPR models. The predictive power of the QSPR model was discussed using a coefficient of determination (R²), average absolute deviation (AAD), mean squared error (MSE), and leave-one-out cross-validation (Q²cv). According to our findings, it was concluded that the QSPR model with three descriptors (ATS5m, BEHm4, and G1) established by the GA-BPANN method could be efficiently used for predicting RI of chemical components in P. minus essential oil and may be helpful for modeling and designing of new molecules of the essential oils.

Keywords:

• QSPR
• Essential Oils
• Back-Propagation Neural Network (BPNN)
• Retention Index
• Molecular Descriptors