Genomic Prediction Ability for Novel Profitability Traits Using Different Models in Nelore Cattle.

Abstract

The aim of this study was to assess the accuracy, bias and dispersion of genomic predictions for accumulated profitability (APF) and profit per kilogram of liveweight gain (PFT) in Nelore cattle using different prediction approaches. The dataset consisted of 3969 phenotypic records for each trait. The pedigree harboured information from 38,930 animals born between 1998 and 2016, including 2691 sires and 19,884 dams. A total of 2449 animals were genotyped using the Clarifide Nelore 3.0 SNP panel. Nine models for genomic prediction were evaluated: a linear animal model was applied to estimate genetic parameters and perform the genomic single-trait best linear unbiased prediction (ST_ss-default). Additionally, a two-trait (ssGBLUP TT_W450 and TT_DMI), three-trait (TTT_CAR) and multi-trait ssGBLUP (MT_ss) were tested. Finally, two models employing the weighted linear (ST_sswl1 and ST_sswl2) and non-linear (ST_sswnl1 and ST_sswnl2) single-step genomic approach (WssGBLUP) were used to predict genomic breeding values (GEBV). The ability to predict future performance was assessed by calculating the correlation between GEBV and adjusted phenotypes. The average prediction accuracy of the GEBV models ranged from 0.345 to 0.665 for PFT and from 0.425 to 0.603 for APF. The predictive capability of the MT_ss model (0.665) was significantly higher than that of the other models for PFT, except for the TTT_CAR model (0.604), which also showed an improvement in predictive performance. For APF, the MT_ss (0.561) and TT_W450 (0.556) models demonstrated improved genomic prediction accuracy compared to the other models. In general, the single trait ssGBLUP (ST_ss-default) models and the non-linear weighting approach did not enhance prediction accuracy for either trait. For the phenotypic prediction ability of PFT, the linear WssGBLUP models ST_sswl1 (0.65) and ST_sswl2 (0.70), TT_W450 (0.64) and ssGBLUP-M (0.66) demonstrated the highest prediction accuracies. Similar results were observed for the phenotypic prediction ability of APF for both models. However, the linear WssGBLUP models ST_sswl1 (0.84) and ST_sswl2 (0.94) provided higher prediction performance compared to the two-, three- and multi-trait models. The results indicate that the multi-trait model achieved better predictive ability for the novel traits PFT and APF. Multi-trait genomic selection may yield greater genetic gains than other models for these forthcoming economically important traits in breeding programmes.