| Resumo: | Sugar analysis contributes to the assessment of their impact on the human health and their physiological effects, allowing to better understand their relation with sensory attributes and acting on quality control and authenticity of food products [1,2]. Although, several analytical methods are routinely used in the identification and quantification of sugars in foods, in general, these methods have several disadvantages such as, slowness of the analysis, high consumption of chemicals and the need for destructive pretreatments of samples. The development of new reliable methods have been proposed [3] to avoid theses disadvantages and, in this follow-up, it was decided to apply a potentiometric electronic tongue, built with cross-selectivity polymeric sensors that were selected considering the sensitivities towards sugars, previously reported [4]. The analysis of sugars (glucose, fructose and sucrose) in this study aimed to establish an analytical methodology and mathematical framework to quantify these compounds. For this purpose, analyzes were performed using standard solutions of ternary mixtures of these sugars, by applying an orthogonal experimental design to establish different concentration levels [5]. It was then made an exploratory data analysis using principal component analysis to verify data variability. To establish a multiple linear relationship between the concentration of sugars and the potentiometric signals obtained by the electronic tongue, a genetic algorithm was used to select the best subset of sensors and cross-validation with K-folds, to optimize the model in prediction. Satisfactory results were obtained in each sugar analysis. For instance, the multiple linear regression model for fructose analysis allowed to have, by cross-validation using K-folds (dividing analytical data randomly into 7 groups), a R²ajusted above 0.99 and RMSE less than 0.5. Moreover, the linear relationship between the predicted values by the obtained model and the respective fructose experimental values allowed to obtain a slope of 0.98±0.02 (close to unity) and an intercept value statistically equal to zero. The multisensor system used proved to be a suitable tool for the analysis of sugars, when present in majority concentrations and alternative to the instrumental reference methods, such as HPLC. It allowed to decrease the time and price of each analysis, and also, to reduce sample preparation work and eliminate pollutants in the analysis procedure.
|
|---|