PSII-4 Genomic prediction for meat tenderness in Nellore cattle: Multi-trait approaches and weighted best linear unbiased predictor.

Abstract Genomic prediction has emerged as a strategic tool for the genetic selection of multifactorial traits that are expensive and difficult to measure, such as meat tenderness. The study aimed to evaluate the impact of different genomic prediction approaches on the prediction ability for meat tenderness in Nellore cattle. Phenotypic data (n = 73,286), pedigree (n = 4,141,892), and genomic data (n = 15,300) from Nellore cattle participating in the genetic improvement program of the National Association of Breeders and Researchers (ANCP) were used. For meat tenderness, phenotypic data from 1,164 animals born between 2002 and 2022 were used. These animals were slaughtered at an average age of 25.55 ± 3.90 months. Six models were tested: (1) ssGBLUP (Single-step Genomic Best Linear Unbiased Prediction), considering the additive direct and residual genetic effects as random effects, with slaughter age as a linear and quadratic covariate (default); (2) Model 1 + ssGBLUP weighted (1st iteration of WssGWAS - Weighted single-step Genomic Best Linear Unbiased Prediction); (3) Model 1 + ssGBLUP weighted (2nd iteration of WssGWAS); (4) Model 1 + weight as a covariate; (5) Model 1 with a bi-trait approach including weight at weaning (W450); (6) Model 1 with a multi-trait approach including carcass traits: rib eye area (REA), backfat thickness (BF), and rump fat thickness (RF). Prediction ability was evaluated using linear regression, dividing the dataset into complete (n = 1,164) and partial (n = 374) subsets. Validation was performed by testing the ability to use marker effects in the training population (older animals with more data) to estimate genomic breeding values (GEBVs) for younger animals. Accuracy values of 0.337, 0.044, 0.044, 0.325, 0.323, and 0.367 were observed for models 1, 2, 3, 4, 5, and 6, respectively. Model 6, which used a multivariate approach, presented the highest accuracy (0.367), suggesting that including these variables improves the capture of genetic variation associated with meat tenderness. Bias was relatively low for all models, ranging from -0.001 (models 2 and 3) to -0.174 (model 4). After models 2 and 3, model 6 showed the lowest bias (-0.105), demonstrating greater reliability. Dispersion ranged from 0.872 (model 1) to 0.952 (models 2 and 3), and correlation varied from 0.838 (model 1) to 0.897 (model 6). Considering all prediction ability parameters, model 6 showed the best performance, with the highest accuracy (0.367), dispersion (0.920), and correlation (0.897), and lowest bias (-0.105), indicating higher efficiency in capturing the genetic variability of meat tenderness, reducing bias, and improving the precision of estimates. Multi-trait models enhance genetic predictions by utilizing trait genetic associations for improved accuracy.