, 2006, Lee and Noble, 2003, Alves et al., 2005 and Kafkas et al., 2006). Nerolidol is a sesquiterpene present in essential oils of diverse plants, showing antibacterial, antifungal and anti-parasite properties (Cowan, 1999). As performed for bitterness, a plot was built of the predicted values
by PLS versus the measured values by QDA (Fig. 5a) and another to evaluate the residuals of the constructed PLS model (Fig. 5b), after GA variable selection. The selected variables for grain taste were well-modelled as can be revealed by the square correlation coefficient, 0.9334, and the root mean square error, 0.27, of the relation shown in Fig. 5a. The residuals (Fig. 5b) were also randomly distributed, confirming the adequate fitting of the selected subset by GA to the grain BMS-754807 supplier taste quality parameter. In relation to OPS variable selection, it was also evaluated the fit among the predicted values by PLS and the measured values by QDA (Fig. 6a). The residuals from this model can be seen in Fig. 6b. The square correlation coefficient was 0.8851 and the root mean square error was 0.25. The correlation coefficient values obtained to the grain taste models can be considered to present an adequate linear relation among the evaluated values since these ones are related to sensorial analysis and the grain taste quality parameter is
not so pronounced as bitterness. As performed to bitterness quality parameter, the variables selected by GA and OPS were evaluated according to its orthogonal behaviour. To verify this occurrence, the correlation coefficient values were obtained among the values selected by GA and
Bcl-w OPS for grain taste, click here as presented in Fig. 7a and b, respectively. It can be seen in Fig. 7a that the GA selected variables presenting low correlation coefficients, indicating that these variables are not correlated between each other. According to Fig. 7b, all the correlation coefficients obtained from the evaluation of the variable selected by OPS presented low values, indicating absence of correlation among them, except to variables 14 and 15. However, these peaks present retention times quite close. Again, according to these results, the genetic algorithm and ordered predictors selection selected basically orthogonal variables, indicating that the useful information is centralised in independent variables. The application of GA and OPS for variable selection allowed the realisation of the correlation between the chromatographic data obtained from 32 commercial beer samples and the data resulting from QDA, for bitterness and grain taste sensorial attributes. The correlation between sensorial and chemical analysis was possible by finding out beer compounds which are linearly related to these quality parameters. The considered substances were that whose peaks were pointed out by both variable selection approaches. The developed PLS models showed the correlation cited above.