Incorporating GO/KEGG Functional Annotations Improves the Accuracy and Stability of Genomic Prediction Across Diverse Beef Cattle Populations

GS in beef cattle faces challenges in cross-population prediction. Blind expansion of reference populations can reduce accuracy because of genetic noise, and traditional models such as GBLUP ignore functional heterogeneity of SNPs. In this study, we used three simulated beef cattle populations with different genetic relationships and a single trait corresponding to birth weight (heritability 0.42) to evaluate the effect of functional annotations on cross-population GS. We compared GBLUP, ssGBLUP and wGBLUP using either all SNPs or SNP sets annotated by GO and KEGG. Mixed reference populations were constructed with multidimensional scaling (MDS)- and fixation index (FST)-based screening. Functionally annotated SNPs increased cross-population prediction accuracy and stability compared with generic SNPs. In the population with close genetic relatedness (PopB), GO-wGBLUP achieved a prediction accuracy of 0.55–0.60 at a 10% reference proportion, higher than GBLUP (about 0.50) and ssGBLUP (about 0.52). In the population with high genetic differentiation (PopA), KEGG-wGBLUP showed a smaller loss of accuracy when the reference proportion increased from 10% to 20% (6.7% decline, from 0.45 to 0.42) than GO-wGBLUP (10% decline, from 0.50 to 0.45) and GBLUP (20% decline, from 0.35 to 0.28). Across scenarios, functional SNP sets reduced the loss of accuracy due to reference population expansion from 20.0% in GBLUP to 12.5% in GO-wGBLUP. These results indicate that wGBLUP combined with GO or KEGG annotations can improve the accuracy and robustness of cross-population GS for beef cattle birth weight traits. GO-based models are more suitable for closely related populations, whereas KEGG-based models are more suitable for highly differentiated populations. The proposed framework provides a practical reference for multi-population GS design in beef cattle breeding.