Abstract
ABSTRACT
Genomic prediction in beef cattle is particularly challenging in breeds with limited phenotypic and genotypic data. In this context, multi‐breed methodologies that integrate information from genetically related populations have emerged as a promising strategy to improve prediction accuracy and calibration under data‐scarce conditions. This study evaluated different genomic prediction approaches in a multi‐breed population of Zebu cattle, including Nellore, Guzerat, Brahman, and Tabapua, using 653 785 phenotypic records, 190 865 genotypic records, and 3 681 158 pedigree records. The traits analyzed were rib eye area (REA), rump fat thickness (RFT), age at first calving (AFC), and accumulated productivity (ACP). Genomic estimated breeding values were obtained using the single‐step GBLUP method under four models: single‐breed (G
SB
), standard multi‐breed (G
0
), metafounders (MF), and an adjusted genomic relationship matrix (AGR). Model performance was assessed using the linear regression method, which compares predictions from complete and partial datasets to estimate accuracy, bias, and dispersion. Multi‐breed models, particularly MF and AGR, produced higher accuracy than the single‐breed approach in several analyses for underrepresented breeds such as Guzerat, Brahman, and Tabapua, especially for carcass and reproductive‐related traits. For example, in Guzerat, REA accuracy increased from 0.44 with G
SB
to 0.62 with AGR, while in Tabapua, ACP accuracy improved substantially from 0.23 with G
SB
to 0.51 with AGR. These results highlight the importance of leveraging genetically related breeds and well‐structured reference populations to improve the reliability of genomic predictions in Zebu cattle under limited data conditions.