Low-field ¹H nuclear magnetic resonance and chemometrics combined for simultaneous determination of water, oil, and protein contents in oilseeds

Abstract

Prediction of the content of water, oil, and protein in rape and mustard seed was examined by a combination of low-field ¹H nuclear magnetic resonance (LF-NMR) and chemometrics, enabling utilization of the entire relaxation curves in the data evaluation. To increase the range of relative contents, the untreated seeds were wetted and dried; each treatment was followed by NMR analysis. The chemometric results are compared to traditional evoluation by multiexponential fitting of the relaxation curves. For this purpose a new jackKnife validation procedure was developed to evaluate the number of exponential components objectively. Classification of the two kinds of seeds was easily performed by LN-NMR. Partial least squares regression to oil content in untreated rape and mustard seed yielded models with correlation coefficients of r=0.88 and 0.89 with root mean square error of cross-validation (RMSECV) of 0.84 and 0.45, respectively. The rapeseed model was based on one component, wheres the mustard seed model was based on two components. If the seeds were dried, the predictive performance improved to r=0.98 and RMSECV=0.38 for rapeseed and to r=0.95 and RMSECV=0.38 for mustard seed. Upon drying, prediction of protein content in mustard seed improved, whereas the prediction of protein for rapeseed deteriorated. Global models, including the combination of untreated, wet, and dry seeds, all resulted in a robust and good predictive performance with RMSECV in the range 0.8–1.3% to water, oil, and protein content. It was demonstrated that drying the seeds to simultaneously determine water and oil content was not necessary when chemometrics was applied on the relaxation curves.