Predicting the composition of red wine blends using an array of multicomponent Peptide-based sensors

Abstract

Differential sensing using synthetic receptors as mimics of the mammalian senses of taste and smell is a powerful approach for the analysis of complex mixtures. Herein, we report on the effectiveness of a cross-reactive, supramolecular, peptide-based sensing array in differentiating and predicting the composition of red wine blends. Fifteen blends of Cabernet Sauvignon, Merlot and Cabernet Franc, in addition to the mono varietals, were used in this investigation. Linear Discriminant Analysis (LDA) showed a clear differentiation of blends based on tannin concentration and composition where certain mono varietals like Cabernet Sauvignon seemed to contribute less to the overall characteristics of the blend. Partial Least Squares (PLS) Regression and cross validation were used to build a predictive model for the responses of the receptors to eleven binary blends and the three mono varietals. The optimized model was later used to predict the percentage of each mono varietal in an independent test set composted of four tri-blends with a 15% average error. A partial least square regression model using the mouth-feel and taste descriptive sensory attributes of the wine blends revealed a strong correlation of the receptors to perceived astringency, which is indicative of selective binding to polyphenols in wine.

Keywords: blends; differential sensing; supramolecular sensors; tannins; wine.

Scheme 1

Schematic representation of the Indicator Displacement Assay (IDA) used for the differentiation of tannins in wine.

Figure 1

Linear Discriminant Analysis (LDA) score plot of UV-Vis responses of the sensing array to the Cabernet Sauvignon/Merlot blends (A), Merlot/Cabernet Franc (B), Cabernet Sauvignon/Cabernet Franc (C) and tri-blends (D). The letters represent the base wines, while the number represent the percentage of each wine. For example, MS28 is a blend of 20% Merlot and 80% Cabernet Sauvignon.

Figure 2

Performance of the Partial Least Squares Regression (PLSR) model using the di-blends as the training set ($●$) in predicting the percentage of Cabernet Sauvignon (A), Merlot (B) and Cabernet Franc (C) in the tri-blend test set ($■$). The calculated RMSEP for the model was 0.18, 0.15 and 0.16 for C. Sauvignon, Merlot and C. Franc, respectively.

Figure 3

PLSR analysis showing the responses of individual receptors in mono varietals. The first three letters of the sensors represent the first letters of the indictor, metal and peptide, respectively. The number represents the wavelength at which the measurement was taken. For example, BNR 560 represents the ensemble BPR/Ni²⁺/RN8 measured at 560 nm.

Figure 4

PLSR correlation plot of the predicting variables in black (responses of the peptide assay) and the predicted variables in red (wine sensory attributes) on the first three dimensions; component 1 vs. component 2 (A) and component 1 vs. component 3 (B). As shown, the peptide sensors are mostly correlate to taste and mouth feel sensory attributes of red wine and, more specifically, to perceived astringency (astrMF).