Using deregressed proofs to estimate multivariate variance components for traits originating from heterogeneous data sources

Abstract

The aim of this study was to investigate the use of deregressed proofs (DRP) for estimating variance components (VC) for traits originating from heterogeneous data sources. Including a Saved Feed index into total merit index (TMI) improves feed efficiency, but TMI weights may need adjustment to avoid undesirable correlated responses. Updating TMI weights requires multivariate VC estimates for the correlated traits. Estimating all VCs in a single analysis may not be feasible when traits differ in data structure, whereas using DRP could offer a practical alternative. The traits considered were grouped into two field datasets (DAT1 and DAT2) and a research-farm dataset (DAT3). DAT1 included milk yield (MY), protein yield (PY), fat yield (FY), cow carcass weight (CCW), cow carcass conformation (CCC), cow carcass fatness (CCF), stature (ST), and metabolic body weight (MBW); DAT2 included bull daily carcass gain (DCG), bull carcass conformation (BCC) and bull carcass fatness (BCF); and DAT3 included MY, PY, FY, MBW, dry matter intake (DMI), and body weight gain (BWG). Observations were from 18,757 cows, 45,514 bull calves and 898 cows for DAT1, DAT2, and DAT3, respectively. After pruning, pedigree sizes were 172,480 for DAT1 and DAT2, and 3,823 for DAT3. For each trait group, we ran a multivariate REML analysis using a repeatability animal model with trait-specific fixed and random effects. In the second part, DRPs were developed for the models used in part one. Reliabilities of breeding values (BVs) for animals with observations were used to derive multi-trait effective record contributions (ERC), which were then used to deregress the BVs with the Broyden method. To validate the use of DRPs in VC estimation, VCs were re-estimated by replacing phenotypic observations with DRPs and fitting a simple animal model with a general mean, a random animal effect, and ERC as weights. The originally estimated heritabilities from the field datasets (DAT1 and DAT2 ) MY, PY, FY, CCW, CCC, CCF, ST, MBW, DCG, BCC and BCF were 0.29±0.02, 0.20±0.02, 0.21±0.02, 0.46±0.03, 0.24±0.03, 0.27±0.02, 0.55±0.03, 0.65±0.03, 0.34±0.03, 0.19±0.02 and 0.20±0.02, and those from DAT3 for MY, PY, FY, MBW, DMI and BWG were 0.46±0.12, 0.28±0.11, 0.25±0.09, 0.84±0.11, 0.27±0.10, and 0.22±0.10, respectively. The heritability estimates and correlations obtained using DRPs were in good agreement with the original estimates when considering their standard errors. Differences between genetic correlations ranged from -0.09 to 0.09. Current results indicate that using DRPs can serve as an alternative for multivariate VC analysis when data sources are diverse, although some deviation from the original estimates should be expected.