QSPR modeling is able to predict retention times of fatty acids using simple molecular descriptors

Abstract

A QSPR modeling was carried out to predict the chromatographic retention times of a series of fatty acid methyl esters (FAME) widely used as standard in the characterization of lipids from agricultural and food products. Number of carbons, total double bonds, position of double bonds and geometric isomerism were used as descriptors to generate a Multiple Linear Regression (MLR) model. The best model yielded an RMSE = 0.167 and R2 = 0.999 for the calibration set, and RMSE = 0.151 and R2 = 1.000 for the test set. Number of carbons and total double bonds were the most important descriptors, according to the regression coefficients, but position of double bonds and isomerism cannot be neglected as they provide relevant information to improve the accuracy of the predicted property.