Prediction of capillary gas chromatographic retention times of fatty acid methyl esters in human blood using MLR, PLS and back-propagation artificial neural networks

Abstract

Quantitative structure–retention relationship (QSRR) models correlating the retention times of fatty acid methyl esters in high resolution capillary gas chromatography and their structures were developed based on non-linear and linear modeling methods. Genetic algorithm (GA) was used for the selection of the variables that resulted in the best-fitted models. Gravitational index (G2), number of cis double bond (N cDB) and number of trans double bond (N tDB) were selected among a large number of descriptors. The selected descriptors were considered as inputs for artificial neural networks (ANNs) with three different weights update functions including Levenberg–Marquardt backpropagation network (LM-ANN), BFGS (Broyden, Fletcher, Goldfarb, and Shanno) quasi-Newton backpropagation (BFG-ANN) and conjugate gradient backpropagation with Polak–Ribiére updates (CGP-ANN). Computational result indicates that the LM-ANN method has better predictive power than the other methods. The model was also tested successfully for external validation criteria. Standard error for the training set using LM-ANN was SE = 0.932 with correlation coefficient R = 0.996. For the prediction and validation sets, standard error was SE = 0.645 and SE = 0.445 and correlation coefficient was R = 0.999 and R = 0.999, respectively. The accuracy of 3–2–1 LM-ANN model was illustrated using leave multiple out-cross validations (LMO-CVs) and Y-randomization.