Visualizing indirect correlations when predicting fatty acid composition from near infrared spectroscopy measurements

Abstract

In recent years, vibrational spectroscopy has been used to predict detailed sample composition like protein and fatty acid profiles. This study shows that fatty acid predictions from near infrared measurements in food stuffs rely on covariance structures amongst the fatty acids. These covariance structures, in turn, vary with factors like breed, age, feed, season etc. and therefore they are not likely to remain constant. Consequently, the robustness and validity of the developed calibration models will be compromised.