Genetic parameters and genomic predictions for economically important traits in Montana® composite cattle using different models

This study assessed the impact of breed effects, heterosis, and recombination on genomic predictions for growth, reproduction, and body composition traits in Montana® cattle. The database included 124,547 records for Birth Weight (BW), 111,103 for Weaning Weight (WW), 87,740 for Weight at 12 months (W12), 49,249 for Post-weaning Weight Gain (WG), 87,740 for Scrotal Circumference (SC), and 44,873 for Muscularity (MUSC), with 3911 genotyped animals from the Montana® composite program. Models M1 to M5 included fixed effects of contemporary group, embryo transfer, and cow age at calving (linear and quadratic). The effects of direct and maternal breed composition, heterozygosity, and recombination varied across models. From model M2 onward, covariates for biological type, heterosis (direct, maternal, specific), and recombination (direct, maternal, specific) were added. All genomic analyses used the ssGBLUP method, and the LR (Linear Regression) validation method was used to assess predictive ability and model effect influence. Heritability estimates ranged from 0.19 to 0.22 for WW, 0.15 to 0.20 for WG, 0.36 to 0.37 for BW, 0.23 to 0.29 for W12, 0.28 to 0.29 for SC, and 0.17 to 0.19 for MUSC. The most parameterized models showed the best fit by AIC, with M5 best for WW, W12, WG, and SC; M4 for BW; and M3 for MUSC. Model M1 showed the best prediction ability for WW and W12, with the highest accuracies (0.407 and 0.456), best dispersions (1.01 and 0.897), and lowest biases (0.098 and 0.068), respectively. For WG, M1 had the highest accuracy (0.452), M5 the best dispersion (0.940), and M4 the lowest bias (0.028). For BW, M5 showed the highest accuracy (0.452), M4 the best dispersion (1.001), and M3 the lowest bias (-0.006). For SC, M1 had the highest accuracy (0.501), M3 the best dispersion (1.004), and M4 the lowest bias (0.092). For MUSC, M4 had the highest accuracy (0.415) and lowest bias (0.057), while M2 showed the best dispersion (0.979). More parameterized models provided a better fit for variance component estimation. In general, genomic predictions with M1 displayed the highest accuracies for WW, W12, WG, and SC, and lower bias for most traits.