Chemometrics in Food Analysis
Data analysis has become a fundamental task in analytical chemistry and in particular in food analysis due to the extensive information provided by nuclear magnetic resonance (NMR) spectroscopy. The chemical composition as determined by quantitative NMR or the NMR spectral fingerprint for a large number of food samples is used as input variables to multivariate statistical analysis or chemometrics, either to unravel natural sample clustering or to establish a classification/prediction model. This approach, which combines the search for metabolites by NMR with statistical analysis, is the emerging field of metabonomics. This chapter presents, briefly, a few practical considerations regarding sample selection and data pre-processing that are necessary steps to produce reproducible input data for statistical analysis. The two metabonomic methodologies, the targeted profiling and chemometric approach for the NMR data manipulation are described concisely giving some practical recommendations for their use. A concise presentation of the standard supervised and unsupervised pattern recognition methods is given with a few pertinent examples, and a section is devoted to the validation of chemometric methods, which constitutes an important aspect of pattern recognition. Finally, a decision tree is afforded as an aid for the choice of the pattern recognition method, which is suitable for a given analysis.