Identification of key mid-infrared wavelengths associated with red deer milk composition : a thesis presented in partial fulfilment of the requirements for the degree of Master of Science in Animal Science at Massey University, Palmerston North, New Zealand

Abstract

Mid-infrared (MIR) spectroscopy has been applied to predict composition of milk from different animal species, but no study has reported the key mid-infrared wavelengths associated with the concentrations of main milk components. The objective of this study was to identify the MIR wavelengths most strongly associated with each milk component to improve understanding of the spectral regions contributing to prediction accuracy in deer milk. A total of 202 milk samples collected from lactating red deer hinds at the Aratiatia Deer Milk Farm, Wairakei, Taupō, New Zealand, were analysed using a MilkoScan 4000 MIR spectrometer. Partial Least Squares Regression (PLSR) models were developed using MIR spectral data to perform an external validation of the equations already integrated in the MilkoScan 4000 MIR spectrometer machine. Model performance was assessed using calibration, 10-fold cross validation, and independent hold-out validation datasets based on the coefficient of determination (R²), root mean square error of prediction (RMSEₚ), and residual predictive deviation (RPD). PLSR coefficient profiles were examined to identify the most influential spectral regions contributing to prediction accuracy. The PLSR models achieved excellent agreement with the FOSS-derived reference values for all major milk components. Validation models produced R² values of 0.9994 for fat, 0.9975 for protein, and 0.9953 for lactose, with corresponding RMSEₚ values of 0.0405%, 0.0282%, and 0.0281%, respectively, demonstrating close agreement between predicted and reference values. The influential spectral regions corresponded closely with established absorption bands associated with lipid, protein, and carbohydrate functional groups, indicating that the models captured chemically meaningful information rather than random statistical relationships. This study provides a greater understanding of the spectral basis of these predictions supporting the development of herd testing systems, genetic improvement programmes, and the emerging deer dairy industry.