Assessment of Mixtures by Spectral Superposition. An Approach in the Field of Metabolomics.

Abstract

A recent field of metabolomic applications is the analysis of mixtures, for example, in food science or the early recognition of diseases. Particularly in large-scale studies, the number of intermediate states or mixtures tends to expand significantly and in practice, a manual analysis is extremely difficult if not nearly impossible. In this study, we present a model in which the NMR spectra of mixtures are calculated based on the spectral superposition of corresponding pure samples. Instead of using real spectra, where chemical shifts may be influenced by matrix effects, the linear combination of reduced data (buckets) was applied for the calculation. Starting from a set of 262 hazelnut samples of five Eurasian countries we obtained more than 160 000 NMR spectra with mixed geographic origin. Using these as a basis we calculated assessment curves to estimate to which extent admixtures are recognized within a multivariate classification model. Subsequently the calculated data were compared with the measured spectra of tangible mixtures to validate and assess the suitability of this method. The calculated spectra are very similar to the acquired data, and the resulting deviations are on a similar scale to the errors of current metabolomic measurements. Thus, with a suitable sample basis, various different mixtures can be simulated and limitations of the model can be described. This approach reduces time and resource consumption and allows valid predictions based on calculated NMR spectra. In addition to the first example dealing with the admixtures classification of single commodity foods, this approach may also be applied to simulate metabolic progression in other areas.