Accuracy of whole-genome prediction using a genetic architecture-enhanced variance-covariance matrix.
Bottom Line: Predictive ability of BLUP/GA was validated with three model traits in a dairy cattle dataset and 11 traits in three public datasets with a variety of genetic architectures and compared with GBLUP and other approaches.Further analyses showed that the difference of accuracies for BLUP/GA and GBLUP significantly correlate with the distance between the T: and G: matrices.Applying BLUP/GA in WGP would ease the burden of model selection.
Affiliation: National Engineering Research Center for Breeding Swine Industry, Guangdong Provincial Key Lab of Agro-animal Genomics and Molecular Breeding, College of Animal Science, South China Agricultural University, Guangzhou 510642, China Department of Animal Sciences, Animal Breeding and Genetics Group, Georg-August-Universität Göttingen, Göttingen 37075, Germany.Show MeSH
Mentions: We first validated BLUP/GA in the dairy cattle population with 5024 bulls (Table 1) via two validation procedures. For the three validated traits (FP, MY, and SCS), the accuracy and unbiasedness of BLUP/GA and GBLUP, obtained from the mean of 20 replicates of fivefold cross-validation (training stage), are shown in Figure 1 and Table 2. The accuracy of BLUP/GA and GBLUP in the candidate population calculated from the application stage and parameters for weight, top% and nflank used for the application stage and derived from the training stage are shown in Table 3. For scenario N = 5024, i.e., all available individuals were used in the training stage, only the optimal parameters derived from the reference population are reported. In both validation procedures, the trend and size of the advantage of BLUP/GA over GBLUP is consistent. BLUP/GA dominated GBLUP for FP and MY, and the advantage decreased with an increased size of the reference population (Table 2 and Table 3), which suggests that using “correct” prior information is particularly important for small datasets, as also noted by de los Campos et al. (2013a). By using BLUP/GA with the optimal set of parameters in a small population (N = 125), the accuracies of genomic prediction were increased by 82.2% and 23.8% in the training stage, and 113.4% and 30.8% in the application stage, for FP and MY, respectively (Table 2 and Table 3). An advantage of 5.6% and 1.9% in the training stage could still be observed for the two traits even if we used the whole population (N = 5024, Table 2). BLUP/GA did not consistently outperform GBLUP for SCS in any of the scenarios investigated (Figure 1, Table 2), which suggests that a zero weight should be assigned to S for this trait in BLUP/GA. With respect to the unbiasedness, both BLUP/GA and GBLUP performed well for all scenarios with this dataset. The optimal parameters applied to BLUP/GA are provided in Table 3. Generally, the optimal S matrix in this dataset was built with a small proportion of top SNPs and five adjacent SNPs on right and left, respectively. While for all traits between 0.01 and 0.5% of the top SNPs were accounted for—with a tendency toward a smaller proportion in smaller reference sets—the average optimal weight assigned to those selected SNPs was 32, 4.0, and 2.0% for FP, MY, and SCS, respectively (Table 3).
Affiliation: National Engineering Research Center for Breeding Swine Industry, Guangdong Provincial Key Lab of Agro-animal Genomics and Molecular Breeding, College of Animal Science, South China Agricultural University, Guangzhou 510642, China Department of Animal Sciences, Animal Breeding and Genetics Group, Georg-August-Universität Göttingen, Göttingen 37075, Germany.