Abstract:
Milk is a very important food resource both in terms of human nutrition and from an economic
standpoint. It is a complex biological fluid containing a huge variety of nutrients; the major
components being water, fats, proteins, and carbohydrates. Milk is valued based on its
composition but methods of determining the composition are often complicated, timeconsuming,
and expensive. Our aim is to use Raman spectroscopy to accurately quantify the
major milk components in goat milk.
Raman spectroscopy has not been widely used in quantification of major milk components but
is a promising technique as it is non-destructive, and has potential for on-farm measurement of
milk. By combining Raman spectroscopy with multivariate data analysis techniques we can
investigate the composition of milk individually from a large number of goats. Changes in the
milk composition of each of the 240 goats from five different farms were investigated using
Principal component analysis (PCA) and Partial least squares (PLS) regression.
PCA was used to explore the Raman spectra of goat milk samples and identify sources of
variation between them. The Raman spectra were shown to be dominated by the variations in
fat and total solids content in the milks. PCA also demonstrated that different farms show small
differences, with milk samples from Farm 4 being higher in unsaturated fatty acids relative to
the other farms.
PLS regression was also used to build calibration models to accurately quantify four major
milk components; fat, protein, lactose, and total solids. The PLS calibration models gave
excellent accuracies for predicting unknown samples, with root mean square errors of
prediction (RMSEPs) for milk components ranging from 0.06 to 0.23 g/100 mL. The PLS
calibration models also achieved residual predictive deviations (RPDs) ranging from 2.8 to 3.8.
Based on the RMSEPs alone the lactose calibration model gave the most accurate
measurements, followed by protein, fat, and total solids.