Fast determination of anthocyanins in red grape musts by Fourier transform mid-infrared spectroscopy and partial least squares regression

The viability of using the Fourier transform mid-infrared spectroscopy matched with partial least squares regression (PLS-R) for the determination of 12 anthocyanins (five non-acylated, three acetylated, three p-coumaroylated and one caffeoylated 3-O-glucosides) in red grapes musts has been studied. Four different PLS regression models were built using 257 samples of grape musts of three harvests, and the prediction of the anthocyanin concentrations using these models was evaluated by internal and external validation sample sets. Cross-validation errors between 9.4 and 30.6 % were attained for major anthocyanins. The major one, malvidin-3-O-glucoside (range of concentrations between 58.78 and 202.76 mg/L), showed standard errors of calibration between 9.01 and 21.12 mg/L. Standard prediction errors for an external validation set including must samples of a new harvest were quite high (between 9.55 and 56.16 %), but somewhat lower (14.91-40.75 %) when considering the regression model including musts of the three harvest. However, good efficiency was observed when using predicted values to order the different musts according to increasing anthocyanin concentration. A correction of FT-IR predicted values should be introduced in order to have more exact absolute values. The proposed method is quick and simple and can be easily implemented for routine winery control of musts and to be used as a quality parameter of harvested grape in terms of their potential contribution to wine color.