Infrared Predictions Are a Valuable Alternative to Actual Measures of Dry-Cured Ham Weight Loss in the Training of Genome-Enabled Prediction Models

Simple Summary Excessive ham weight losses during dry-curing (WL) result in a loss of marketable product, hindering the quality of dry-cured hams. Genetic selection for reducing WL requires individual traceability of hams throughout the dry-curing process and the measurement of WL is expensive, time-consuming, and can be performed only at the end of seasoning, resulting in long generation intervals. Infrared spectroscopy provides early, cost-effective, high-throughput predictions of WL that are highly genetically correlated with the actual measures. This study focused on the accuracy of genomic prediction models for observed and infrared-predicted WL. Models were tested on crossbred pigs and their purebred sires in random cross-validation and in a leave-one-family-out training-validation scheme. Accuracy of prediction of sire genetic merit, estimated from crossbred training data for actual ham WL, was 0.38, slightly higher than the accuracy attainable by a model trained on infrared-predicted WL (0.32). While the accuracy of genomic predictions is satisfactory for both the observed and infrared-predicted WL, the use of infrared predictions results in considerably lower phenotyping costs, enabling the construction of larger reference populations. Selection to reduce ham weight losses during dry-curing (WL) requires individual traceability of hams throughout dry-curing, with high phenotyping costs and long generation intervals. Infrared spectroscopy enables cost-effective, high-throughput phenotyping for WL 24 h after slaughter. Direct genomic values (DGV) of crossbred pigs and their purebred sires were estimated, for observed (OB) and infrared-predicted WL (IR), through models developed from 640 and 956 crossbred pigs, respectively. Five Bayesian models and two pseudo-phenotypes (estimated breeding value, EBV, and adjusted phenotype) were tested in random cross-validation and leave-one-family-out validation. The use of EBV as pseudo-phenotypes resulted in the highest accuracies. Accuracies in leave-one-family-out validation were much lower than those obtained in random cross-validation but still satisfactory and very similar for both traits. For sires in the leave-one-family-out validation scenario, the correlation between the DGV for IR and EBV for OB was slightly lower (0.32) than the correlation between the DGV for OB and EBV for OB (0.38). While genomic prediction of OB and IR can be equally suggested to be incorporated in future selection programs aiming at reducing WL, the use of IR enables an early, cost-effective phenotyping, favoring the construction of larger reference populations, with accuracies comparable to those achievable using OB phenotype.