Near infrared hyperspectral data were collected for 200 Syrah and Tempranillo grape seed samples. Next, a sample selection was carried out and the phenolic content of these samples was determined. Then, quantitative (modified partial least square regressions) and qualitative (K-means and lineal discriminant analyses) chemometric tools were applied to obtain the best models for predicting the reference parameters. Quantitative models developed for the prediction of total phenolic and flavanolic contents have been successfully developed with standard errors of prediction (SEP) in external validation similar to those previously reported. For these parameters, SEPs were respectively, 11.23 mg g-1 of grape seed, expressed as gallic acid equivalents and 4.85 mg g-1 of grape seed, expressed as catechin equivalents. The application of these models to the whole sample set (selected and non-selected samples) has allowed knowing the distributions of total phenolic and flavanolic contents in this set. Moreover, a discriminant function has been calculated and applied to know the phenolic extractability level of the samples. On average, this discrimination function has allowed a 76.92% of samples correctly classified according their extractability level. In this way, the bases for the control of grape seeds phenolic state from their near infrared spectra have been stablished.
Keywords: chemometrics; extractability; flavanols; grape seeds; near infrared; phenolic compounds; total phenols; vibrational spectroscopy.