Non-targeted and targeted analytical approaches for estimating the features of wine spirits

Abstract

Non-targeted UV–Vis/synchronous fluorescence (SF) spectroscopic and differential pulse voltammetric (DPV) fingerprinting have been tested to discriminate between caramel-containing and caramel-free wine spirits using partial least squares-discriminant analysis (PLS-DA), variable importance in projection (VIP)-PLS-DA, VIP-principal component analysis-linear discriminant analysis (PCA-LDA) and VIP-support vector machine (SVM). A 96 % total correct classification in VIP-PLS-DA and VIP-PCA-LDA was achieved using, among others, of 115 variables corresponding to a potential range from 0.43 to 1.00 V in DPV, or 45 variables selected from SF spectra recorded at a difference between the emission and excitation wavelengths of 100 nm for diluted samples. The conclusions resulting from the non-targeted methods were verified and confirmed by the targeted quantitative analysis of the samples using the HPLC-DAD method, focused on 5-hydroxymethylfurfural, furfural, syringaldehyde and vanilline. In targeted SF spectroscopic quantitative analysis, the PLS regression model based on 21 variables selected from SF spectra recorded at a difference between the emission and excitation wavelengths of 100 nm was suitable for the determination of caramel in the range from 0.1 to 4.0 g/L. High coefficients of determination R2 and small root mean squared error (RMSE) values for both cross-validation and prediction (R2 = 0.9811 and 0.9736, respectively, and RMSE = 0.046 and 0.047, respectively) prove good quantitative properties of the VIP-PLS regression model.