Desenvolvimento de método quimiométrico baseado em análise de imagens digitais para a quantificação dos corantes amarelo crepúsculo e tartrazina em bebidas

Abstract

The main goal of this dissertation was the development and validation of a chemometric methodology for the simultaneous determination of two artificial food dyes, sunset yellow (SY) and tartrazine (TA), in non-alcoholic beverages, such as soft drinks, isotonics and artificial juices. For this goal, it was necessary a previous optimization of a high performance liquid chromatography (HPLC) method and its validation in order to obtain reference values for the multivariate calibration method. The HPLC method was developed using a C-18 column, a mobile phase composed of acetonitrile/methanol 80:20 (v/v) and ammonium acetate (pH = 6.7) solution, and spectrophotometric detection. A number of 123 different samples of several types, brands and flavors were analyzed. Quantitative models were built with the reference HPLC values, digital image analysis and chemometric tools, such as PLS (Partial Least Squares) and OPS (Ordered Predictors Selection). Digital Images were generated with a simple scanner, using ultrasonic bath as the only sample pretreatment. RGB histograms obtained from digital images were used as analytical signals. The analytes were determined in the range from 2.3 to 41.1 and from 0.1 to 15.1mg.L-1, for SY and TA, respectively. The best PLS models provided RMSEP (root mean square error of prediction) of 2.8 and 2.6 mg.L-1 for SY and TA, respectively. A variable selection with OPS allowed reducing the number of variables used in models construction from 768 to 100, providing predictions similar to the previous models, with RMSEP of 2.6 and 2.7 mg/L for SY and TA, respectively. All the models were validated through the estimate of appropriate figures of merit. The developed chemometric methods were rapid, of low cost, requiring a minimum sample pretreatment, and clean, not consuming reagents nor generating residues.