This exploratory study aimed to classify Portuguese Protected Designations of Origin wine grape must varieties using supervised models (LR, SVC, PLS-DA, GBC and RF), and model interpretation by computing Shapley values to perform feature analysis, by reusing the available data from Fourier-Transform Infrared spectroscopy. The different variety models reached a range from 0.69 to 0.98 of area under the Precision Recall curve. The specific features with highest impact on each variety model output were identified, and allowed the discrimination of grape must varieties. Aragonez grape must presents two non-correlated features, the bands at 1127 cm \(^{-1}\) and 1130 cm \(^{-1}\) , related to alphatic compounds. Syrah grape must has one feature, the band at 1161 cm \(^{-1}\) , associated to strong vibration in acyl groups of esters compounds. Touriga Nacional grape must possesses three non-correlated features, related to monosubstituted alkenes, the bands at 999 cm \(^{-1}\) , 1015 cm \(^{-1}\) and 1123 cm \(^{-1}\) . The white grape must varieties, Arinto and Viosinho, have also a unique feature, the band at 2924 cm \(^{-1}\) , related to carbon dioxide compounds with a O=C=O bonds pattern vibrations. Our results highlight the contributions of wine grape must metabolomics data to the authentication process of international and Portuguese native wine grapes.