Fig. 1: Dimensionality reduction of chromatograms reflects the geography of the Bordeaux region.

a t-SNE plot of the 80 concatenated chromatograms with the first two embedding dimensions. Colors correspond to different estates while vintages appear next to each data point. The resulting map recapitulates the geography of the Bordeaux region up to a rotation, but note that the overall orientation of the t-SNE projection is arbitrary. Wines from the same estate but different vintages tend to cluster together, with little overlap between clusters. Right bank estates (A, C and B from Pomerol and St-Emilion) and left bank estates (F, G, D, and E from Medoc) also tend to cluster together. Moreover, the left bank estates are organized along a north-south axis, E and D being the furthest north, while G and F are closer to Bordeaux. b Same as (a) but with the UMAP algorithm. c The 7 estates in our data set are shown in the same colors as in (a), coming from the two regions in South-West France (inset), highlighted by the ellipses on the right (Pomerol and St-Emilion) and left (Medoc) bank of the Garonne river. The clusters follow the same general organization as for t-SNE. d t-SNE applied to varietal percentages. e same as (d) but with UMAP. In (e), the distinction between right and left bank estates is less clear, and in both (d) and (e), there is no north-south axis on the left bank and some estates (A and B, E and D) are no longer distinguishable. This suggests that the blend is not the sole contributor to the map obtained with GC.