Comparison of Multivariate Regression Methods for the Analysis of Phenolics in Wine Made from Two Vitis vinifera Cultivars

Abstract

The quantification of key phenolic classes during different stages of red wine production has important industrial applications. Decisions of this nature require greater speed and economy than current bench or instrumental methodologies offer. Efforts to create rapid analysis have used models created from multivariate regression of easily obtained spectra (ultraviolet (UV)/visible or near infrared) and referenced phenolic measurements. For this experiment, reference measures of phenolics and UV/visible spectra were gathered from 100 samples each of Cabernet Sauvignon and Syrah red wine throughout fermentation to create a model that could rapidly predict several phenolic classes in red wine must. The reference method, UV/visible spectra sample dilution, and multivariate regression method were all varied to determine which combination gave the greatest predictive power. Ridge regression unanimously outperformed the other regression methods tested when calibrated with the modified assay at pH 7 and UV/visible sample dilution at 10-fold for only the Cabernet Sauvignon samples. Correlation coefficients obtained for anthocyanins, small and large polymeric pigments, tannins, and total iron reactive phenolics were 0.83, 0.78, 0.76, 0.92, and 0.90, respectively. When the same multivariate regression evaluation was performed for Syrah, none of the methods tested gave accurate predictions, suggesting some cultivar specificity.