Abstract
Elucidating the chemistry of wine would help defining its quality, chemical and sensory characteristics and optimise the wine-making processes. High-performance liquid chromatography coupled with UV–Vis spectroscopy (HPLC–UV–Vis) is a common analysis method used to obtain the molecular profile of wine samples. We propose a complete procedure for the analysis of wine chromatograms. Data are pre-processed using standard methods of down-sampling, smoothing and baseline subtraction. Multiple samples are then merged in a three-dimensional tensor, decomposed using parallel factor analysis (PARAFAC2) into three factors: (i) one reduced (rank-one) chromatogram per sample, (ii) an estimate of the samples’ spectral UV–Vis profile and (iii) an estimate of the samples’ concentrations. If the decomposition is performed on a single peak of the tensor, the second and third factors correspond to the representative wavelength spectrum and to the relative concentrations of the samples, respectively. Otherwise, when multiple peaks are analysed, further processing is required. In the latter case, the decomposed rank-one chromatograms are peak-detected and aligned, clustered and integrated. A table containing the concentration of the peaks at different retention times is obtained. The pipeline proposed in this study is a guideline for a quantitative and reproducible chemical analysis of wine, or other samples, via the HPLC–UV–Vis method.