Application of near-infrared spectroscopy/artificial neural network to quantify glycosylated norisoprenoids in Tannat grapes

Abstract

Grape variety, vinification, and ageing are factors conditioning the aroma of a wine, with volatile secondary metabolites responsible for the so-called grape varietal character. Particularly, grape glycosylated norisoprenoids are mostly responsible for the sensory profile of Tannat wines, making relevant the use of fast instrumental tools to evaluate their concentration, allow classifying grapes and defining the optimum maturity for harvest. NIR spectroscopy is a fast, non-destructive technique, which requires minimal sample preparation. However, its quantitative applications need chemometric models for interpretation. In this work, a NIR-ANN calibration was developed to quantify norisoprenoids in Vitis vinifera cv. Tannat grapes during maturation and harvesting. Glycosidated norisoprenoids were determined by GC-MS. The ANN adjustments showed better performance than linear models such as PLS, while the best calibration was obtained by homogenising grape samples when comparing to grape juice; making possible to fit a model with an error of 146 μg/kg.

Keywords: Artificial neural networks; Glycosylated norisoprenoids; Near infrared spectroscopy; Tannat grape.