Abstract
In this paper, Nuclear Magnetic Resonance spectroscopy (NMR)-based metabolomics were applied for the discrimination of ale and lager craft American beers. A modified pulse sequence that allows the efficient suppression of the water and ethanol peaks was used to achieve high-quality spectra with minimal sample preparation. The initial chemometrics analysis generated models of low predictive power, indicating the high variability in the groups. Due to this variability, we tested the effect of various data pretreatment and chemometrics approaches to improve the model's performance. Spectral alignment was found to improve the classification significantly, while the type of normalization also played an important role. NMR combined with statistical and machine-learning techniques such as orthogonal projection to latent structures discriminant analysis (OPLS-DA) and random forest was able to discriminate between ale and lager beers, thus providing an important tool for the quality control and analysis of these products.