Graph theory algorithms, a novel way of exploring red wine 1H-NMR data.

Wine is a highly appreciated product consumed worldwide. It is a beverage with an extensive aroma and taste diversity, stemming from a wide natural variation in its chemical composition. Nuclear magnetic resonance (NMR) is a powerful analytical technique that enables precise analysis of complex chemical compositions offering both qualitative and quantitative capabilities generating a chemical fingerprint of a wine sample. This technique is often combined with multivariate statistical modelling or chemometrics to process a large amount of data for both classification of samples and detection of adulterations. In metabolomics, traditional algorithms based on Principal Component Analysis (PCA) have been employed for data exploration and investigation of trends. However, these algorithms rely on the assumption that the data is normally distributed to perform the dimension reduction, which, for NMR data is not always the case. The same issue is present in other fields, such as genomics and proteomics where new algorithms based in graph theory such as Uniform Manifold Approximation (UMAP) have been employed with great success. This approach allows to both avoid the previously mentioned assumption, while comparing each point of the dataset individually, which might provide a better representation of the NMR data. At the Research Unit “Product Identity, Supply Chains and Traceability” of the German Federal Institute for Risk Assessment (BfR) research on wine chemical composition has been conducted for several years, resulting in the accumulation of a database with thousands of wine 1H-NMR spectra from samples worldwide. This work aims to present an in-house developed data exploration tool using algorithms that are yet to be applied for the study of NMR data (SPIMP, UMAP and HDBSCAN). And apply to large library of red wines available at BfR with the goal of producing a workflow capable of classifying red wine based on regionality and grape variety.