A contribution to the harmonization of non-targeted NMR methods for data-driven food authenticity assessment

Spectroscopic non-targeted methods are gaining ever-growing importance in quality control and authenticity assessment of foodproducts because of their strong potential for identification of specific features of the products by data-driven classifiers. One ofthe factors hampering the diffusion of spectroscopic non-targeted methods and data-driven classifiers is the lack of harmonizedguidelines for their development and validation. In particular, to date, neither conditions to directly compare spectra recorded bydifferent spectrometers nor studies demonstrating the statistical equivalence of the spectra are available.Among the spectroscopicanalytical techniques suitable for the development of non-targetedmethods, nuclear magnetic resonance (NMR) offers the uniqueopportunity to generate statistically equivalent signals. In this paper, the feasibility of NMR spectroscopy to generate statisticallyequivalent NMR signals from a number of different spectrometers was demonstrated for complex mixtures (aqueous extracts ofwheat and flour) by organizing an inter-laboratory comparison involving 36 NMR spectrometers. Univariate statistics along withmultivariate analysis were exploited to establish unbiased criteria for assessing the statistical equivalence of the NMR signals.The aspects affecting the signal equivalence were investigated, and possible solutions to reduce the extent of the human errorwere proposed and applied with satisfactory results. This study furnishes the scientific community with an appropriate and easyprocedure to validate non-targeted NMR methods and provides error values to be used as a reference for future studies.