Abstract
Superiority of genomic selection (GS) is argued to be due to better modeling of the Mendelian sampling (MS) and tracking of QTL similarities between individuals. It is not clear that a better genome-wide modeling of MS contributes to the increased accuracy. In fact, it might be that modeling of MS outside areas of the genome under selection pressure is detrimental to the accuracy of GS. If true, this hypothesis will provide a better framework to understand the complex relationships between MS, QTL similarity and accuracy. Increases in marker density and the need for marker prioritization makes this hypothesis even more practically important. Answering this question could have a significant impact on accuracy and the computational costs of GS implementation. A 30-chromosome genome with 50K SNPs was simulated. 200 QTL were simulated on two chromosomes for a trait with heritability of 0.4. Genomic relationships were calculated based on all 50K SNPs (G30), 3,333 SNPs on the two chromosomes carrying QTL (G2), and 46,667 SNPs on chromosomes without QTL (G28). Table 1 shows accuracies after 3 and 10 generations of (G)EBV-based selection (M1) and random selection (M2). BLUP accuracies are consistently higher (11.5 to 43.8%) than G28, showing that expected relationships better model QTL similarities than a dense panel of markers that lie outside QTL regions. Inclusion of markers that lie outside QTL regions with markers inside QTL regions reduces accuracies, as shown by the inferior (20.2 to 22.8%) performance of G30 compared to G2. Coefficients of variation were higher for low than high additive relationships suggesting that errors made in estimating QTL similarities for lowly related animals may have the most detrimental impact. Furthermore, while G28 markers capture more variation than pedigree, the superiority of BLUP indicates that variation captured by G28 is not consistent with true variation in QTL inheritance.