The multivariate advantage in fat determination in meat by bench-top NMR

Abstract

The intention of the present work was to investigate the potential improvement of a multivariate extension to the existing measurements of fat in meat performed with bench-top NMR. State of the art in measuring fat in raw meat containing water is the application of the gradient stimulated echo experiment to suppress the water signal. In order to improve this univariate experiment (the echo amplitude) a new pulse experiment was designed where a 180°-pulse train (as used in CPMG) was appended at the end of a pulsed field gradient stimulated echo experiment (DIFF-CPMG). The multivariate advantage of this experiment when evaluated by partial least squares regression (PLSR) is clearly demonstrated. When applied to a designed set of 47 raw meat samples which were also analysed by conventional solvent extraction (SBR), the new multivariate field-gradient pulse-sequence yielded a root mean square error of cross-validation (RMSECV) of 0.49% fat to be compared with the prediction error for the gradient-echo experiment of 1.50% fat. Simultaneous prediction of water content with DIFF-CPMG displayed a RMSECV of 0.56. Moreover, this study also reveals that if high prediction precision of fat is required, drying prior to NMR analysis improves the prediction error to RMSECV≈0.25% fat independent of the pulse experiment performed, but at the expense of additional analysis time and sample handling.