PSIV-20 Impact of a response with measurement errors on the joint genetic evaluation with correlated noise-free responses.

Abstract Phenotyping of novel traits has seen spectacular improvement during the last few years. Using automated and “quasi-automated” approaches, trait phenotypes that were previously impossible or costly to measure (e.g., methane emissions, horn fly counts) are being routinely collected. This will undoubtedly enhance selection objectives and improve the productivity and sustainability of the livestock industry. Unfortunately, several novel phenotypes are measured with non-negligible noise (errors). In fact, for some of these traits, the measurement error level can exceed 30%. The impact of this noise on the estimation of genetic parameters and the prediction of breeding values is severe, even at low levels (< 10%). However, the impact of noisy trait phenotypes on the estimation of genetic parameters of noise-free correlated traits included in a multi-trait genetic evaluation is unknown. In this study, two correlated traits (r = 0.10 and r = 0.90) and heritabilities of 0.4 were simulated following a mixed linear model that included one fixed effect with 20 classes and the additive genetic effect. The pedigree file included 101,000 animals. Measurement noise was introduced on one of the two traits using a multiplicative error factor (v). When v was sampled from a uniform distribution between 0 and 0.2 (measurement noise of around 10%) and the true genetic correlation was 0.9, heritability estimates were 0.26 and 0.39 for the noisy and noise-free traits, respectively. Estimated genetic and residual correlations were 0.9 and 0.69, respectively. The Pearson correlation between the true and estimated breeding was 0.74 for the noisy-free trait and 0.50 for the noisy trait. When v was sampled from a uniform distribution between 0 and 0.6 (measurement error of around 30%) and the true genetic correlation was 0.9, the estimated heritability of the noisy trait, the residual correlation, and the Pearson correlation between true and estimated breeding values decreased to 0.05, 0.26, and 0.35, respectively. No change was observed in the estimates of the genetic correlation and the Pearson correlation between true and predicted breeding values for the noisy-free trait. In the weakly correlated trait scenario (r = 0.10), a similar trend for the genetic parameter and accuracy estimates was observed, although the magnitude of the decrease was smaller for the noisy trait. Collectively, these results indicate that measurement errors significantly affect the genetic evaluation of a noisy trait but do not seem to affect correlated responses. However, that might not be the case when measurement errors are different across classes of fixed effects or sire families.