Predictive ability of MIR Spectra for detecting non-coagulating milk of Sarda ewes

Milk coagulation ability is crucial in the dairy industry. Non-coagulating (NC) samples are defined as milk not forming a curd within the testing time (RCT) of 30 min. In Sheep milk, up to 10% NC samples has been reported both in individual and bulk milk. Although the clotting properties of individual milk have been widely studied, little attention has been given to NC milk and these samples are often removed from the analysis. MIR spectra can be exploited both to predict cheese-making aptitude and to discriminate between coagulating and NC samples. Milk sample from 1,018 Sarda ewes from 47 flocks located in Sardinia (Italy) were analysed and served as training dataset (TD). Validation dataset (VD) were of 662 ewes sampled from the same flocks but one year later (1-5 controls) for 2,656 records in total. Three classical MCP were measured: rennet coagulation time (RCT), curd firmness (a30) and curd firming time (k20). MIR spectra were recorded in the region between 925.92 and 5,011.54 cm-1. In order to predict the coagulation status (binary trait: 0=NC, 1=Coagulating): (1) principal component of MIR spectra and logistic regression (PC-LR) on coagulation status; and (2) linear discriminant analysis (DA) were applied. The effect of different MIR regions combinations/exclusion on predictive ability of NC sample was also assessed. About 9.5 and 5.5% did not coagulate at 30 min for TD and VD, respectively. The use of logistic regression on PC extracted from MIR spectra gave the poorer results (correct assignment of NC samples <57% for TD). PC-LR selecting and combining different regions of MIR spectra did not improve the % of correct assignment. DA combined to step-wise variables selection raised up to 87% the correct assignment in TD. As far as external validation concern DA was carried out on VD for different number of test day available (23.1-100% of correct assignment, >90% if at least 3 test day were available per animal).