Discrimination of autochthonous pig breeds from meat and fat samples by FT-NIR spectra

The combination of Fourier transform near-infrared spectroscopy (FT-NIRS) of meat and fat samples and principal component discriminant analysis (DAPC) has been proposed as a tool for discriminating the local production of autochthonous pig breeds. Spectral samples (n = 272) belonging to 11 local European pig breeds, Longissimus muscle, and subcutaneous fat (both intact and minced) are collected. Classification accuracy based on DAPC was applied on FT-NIRS to predict breed of origin in i) semi-external cross-validation, splitting the data into training (80 %) and testing (20 %) sets; ii) external validation, in which one breed at a time was excluded from model training and classified in one of the remaining breeds. The effect of varying sample sizes from 50 % to 100 % of the data was assessed. Almost all breeds' spectra variability was summarised into two principal components for tissue and sample preparation. In cross-validation, intact fat yielded higher classification accuracies than intact meat, with less pronounced differences in minced samples. Success assignment rates of ∼81–83 % were obtained for two breeds in intact meat samples and were higher than 83 % for five breeds in fat samples. For minced samples, correct assignments between 80 % and 100 % were possible for five breeds, both in meat and fat samples. Sample size marginally affected the results. External validation confirmed similarity among some breeds, with greater accuracy for fat samples. The assignments success provides encouraging results for discriminating local pig production, mainly based on fat, using a rapid, eco-friendly FT-NIRS method, which could serve as tool for quality assurance. • FT-NIRS on meat and fat is promising in the discrimination of autochthonous breeds. • Common NIR spectra properties, allowed to summarise almost all of the variability. • Fat performed better results than meat for the breed assignment success. • Differences in sample presentation: lower performance of intact samples.