Influence of variance component estimates on genomic predictions for growth and reproductive-related traits in Nellore cattle

Abstract

This study aimed to estimate variance components (VCs) for growth and reproductive traits in Nellore cattle using two relationship matrices (pedigree relationship A matrix and pedigree plus genomic relationship H matrix), and records collected before and after genomic selection (GS) implementation. The study also evaluated how genomic breeding values (GEBV) are affected by variance components and discarding old records. The analysed traits were weight at 120 days (W120), weight and scrotal circumference at 450 days (W450 and SC450, respectively). Three datasets were used to estimate VCs, including all phenotypic information (All) or records for animals born before or after GS implementation (Before or After datasets, respectively). Both relationship matrices were considered for VC estimation, the A matrix was used in all three datasets and VC from each combination were named as A_Before, A_After, and A_All). The H was used in two datasets: H_All and H_After. Different VCs were used for GEBV prediction through ssGBLUP. This step used two possible Datasets, using all available phenotypic data (Dataset 1) or just records collected since GS implementation (Dataset 2). Validation was conducted using accuracy, bias and dispersion according to the LR method and prediction accuracy from corrected phenotypes. The heritability of all traits increased from A_Before to A_After, while estimates for A_All were intermediary. In the previous order, the estimates were 0.16, 0.17, and 0.15 for W120; 0.31, 0.39, and 0.35 for W450; 0.35, 0.47, and 0.41 for SC. For W450 and SC, using the H matrix reduced the heritability (0.33 and 0.32 for W450; 0.41 and 0.38 for SC for H_After and H_All, respectively). For W120, Dataset1 and VCs from A_After showed the highest accuracy for direct and maternal GEBV (0.953 and 0.868). For W450, Dataset 1 and VC from H_After allowed the highest accuracy (0.854) but use Dataset 2 and same VC source yield similar value (0.846). For SC, Dataset 2 with VC from H_After showed the highest accuracy (0.925). To use Dataset 2 does not cause important changes in bias or dispersion with respect to Dataset 1. The VC and genetic parameters changed for W120, W450, and SC450, using records before or after the GS implementation. For W450 and SC450, genetic variance and heritability estimates increased with the use of GS. For W120, genomic predictions were more accurate using A for VC estimation. Accuracy gains were observed for W450 and SC450 using H in VC estimation and/or discarding records before GS. It is possible to discard phenotypic records before GS implementation without generating bias or dispersion in the GEBV of young candidates.