Determination of Cabernet Sauvignon wine quality parameters in Chile by Absorbance-Transmission and fluorescence Excitation Emission Matrix (A-TEEM) spectroscopy. (30th October 2022)

• Record Type: Journal Article
• Authors: Schober, Doreen; Gilmore, Adam; Chen, Linxi; Zincker, Jorge; Gonzalez, Alvaro
• Is Part Of: Food chemistry. Volume 392(2022)
• Publication Date: 2022-10-30
• Subjects: ANOVA Analysis of variance -- A-TEEM Absorbance-transmission and fluorescence excitation emission matrix -- CV Cross validation -- EEM Excitation emission matrix, HPLC-DAD: High performance liquid chromatography-diode array detection -- IFE Inner filter effects -- LOD/Q Limit of detection/quantitation -- PLSR Partial least squares regression -- RM Rayleigh masking -- RMSE root mean square error -- TPP Total polymeric pigments, XGBR: Extreme gradient boost regression; Fluorescence -- Machine learning -- Phenolics -- Quality markers -- Spectroscopic methods -- Wine industry; Food -- Analysis -- Periodicals; Food -- Composition -- Periodicals; 664
• DOI: 10.1016/j.foodchem.2022.133101
• Languages: English
• ISSNs: 0308-8146

Abstract

Highlights: Basic chemistry and phenolic compounds were associated to Chilean Cabernet Sauvignon wine quality. A-TEEM was newly developed as an efficient quantification technique for wine quality parameters. Absorbance and fluorescence EEM data combined yielded optimum results. The XGBR compared to PLSR algorithm with data pre-preprocessing yielded lowest prediction bias. The calibration models were validated to be consistent with a set of wines from a subsequent harvest. Abstract: A-TEEM spectroscopy is presented as a novel rapid quantitative analysis method for 44 individual phenolic and basic wine chemistry compounds. To date no practical and combined analysis method for these recognized quality parameters important to the wine industry exists. The method was implemented in a Lambert-Beer linear concentration range to facilitate traceable absorbance and fluorescence spectral signatures. Both components were comparatively analyzed as single- and combined multi-block variable sets, and regressed against HPLC–DAD, UV–vis spectroscopy and other analytical reference data, using the Extreme Gradient Boost Regression (XGBR) and Partial Least Squares Regression (PLSR) algorithms. The approach was applied on 126 wines, and subsequently validated by a random split of 13% of the set and an additional independent set of 16 wines. XGBR with multi-block data organization systematically yielded the highest prediction accuracy and precision with respective overall valid fits indicated by mean R 2 and relative bias of 0.94 ± 0.04 and 4.1 ± 1.8%.