Abstract #650

# 650
Use of partial least squares regression to predict individual milk coagulation properties and cheese yield from Fourier transform infrared spectra in Sarda dairy sheep.
Maria Grazia Manca1, Jessica Serdino1, Massimo Cellesi1, Paolo Urgeghe1, Ignazio Ibba2, Marino Contu2, Nicolo P. P. Macciotta*1, 1Dipartimento di Agraria, Università di Sassari, Sassari, Italy, 2Associazione Regionale Allevatori della Sardegna, Cagliari, Italy.

Milk coagulation properties (MCP) are popular indicators of milk cheese making ability. They are measured as rennet coagulation time (RCT, min), curd firming time (k20, min) and curd firmness (a30, mm). The potential cheese yield of milk could be also be assessed by individual cheese micro-manufacturing experiments (ILCY). However, the routine measure of these traits appears to be rather problematic in terms of costs and logistics. In this work, partial least squares regression (PLSR) is used to predict individual MCP and cheese yield of 965 Sarda breed ewes located in 47 flocks. MCP were measured using the Formagraph, and cheese yield was assessed by ILCY. Mid infrared spectra was obtained by Milkoscan (Foss Electric). Animals were split into 2 data sets: (1) training (700 ewes) that was used to estimate the PLS model and (2) validation (265 ewes), that was used to validate PLS predictions. One hundred replicates were performed, randomly assigning animals to training and validation sets. Goodness of predictions was assessed by calculating the determination coefficient (R2), the residual mean squared error of prediction (RMSEP), the regression slope (bobs,pred) and intercept (aobs,pred) (Table 1).The R2 indicates an accurate prediction for RCT and, to a lesser extent, a30, and very poor for k20. Also for ILCY the prediction was quite accurate. These figures were confirmed also by values of RMSEP and of regression parameters. The PLSR yielded prediction results of moderate accuracy for RCT and ILCY using mid-infrared spectral data as predictors and it could represent a valuable tool for the recording of these phenotypes for management and breeding purposes. Research supported by the regione Autonoma della Sardegna, project “Il latte Ovino della Sardegna” Table 1. Statistics of PLS prediction for MCP and ILCY in the validation data set
Mean obsMean predR2RMSEPbobs,predaobs,pred
RCT (min)15.25 ± 6.6215.20 ± 5.820.71 ± 0.053.57 ± 0.350.97 ± 0.080.57 ± 1.17
k20 (min)1.53 ± 0.851.55 ± 0.380.07 ± 0.030.83 ± 0.050.60 ± 0.170.60 ± 0.27
a30 (mm)49.8 ± 20.249.8 ± 15.90.55 ± 0.0513.5 ± 0.80.95 ± 0.082.5 ± 4.4
ILCY (%)36.3 ± 9.336.3 ± 7.70.63 ± 0.065.7 ± 0.50.95 ± 0.061.7 ± 1.93

Key Words: milk coagulation properties, sheep, partial least squares regression