Date of this Version
Including marker-assisted selection in breeding programs is potentially more efficient than traditional selection for improving traits that are expensive or difficult to measure. One of the challenges of genomics is the lack of robustness of marker effects across populations and over time (generations) and the cost to commercial producers of high-density arrays. The objective of this study was to analyze differences in the proportion of phenotypic variation explained by different fractions of major 1 Mb windows and SNPs. Using a population of Nebraska Index Line and commercial Large White x Landrace females (n = 1,234) generated in 11 batches, we conducted a genome-wide association analysis for age at puberty (AP) using a Bayes B algorithm with a π value of 0.99 and the concatenation of diet and batch fitted as a fixed effect. A total of 56,424 SNPs explained 0.28 of the phenotypic variation for AP. Analysis of the genetic variance explained by 1 Mb windows across the genome and major SNPs, uncovered major regions associated with AP. The proportion of the phenotypic variation explained by all SNPs within the top 1%, 5%, 10% and 20% windows varied from 0.22 (1% windows; 645 SNPs) to 0.39 (10% windows; 19,362 SNPs). In contrast, the proportion of the phenotypic variation explained by the most informative SNP from these windows varied from 0.18 (1% windows; 24 SNPs) to 0.48 (20% windows; 259 SNPs). Different π values (0, 0.25, 0.50, 0.75 and 0.99) had a limited effect on the proportion of phenotypic variation explained by the top 1% (0.20 to 0.23) and 10% (0.36 to 0.37) windows. The first seven batches were used as training data (R1 - R7, n = 822) to evaluate the ability of major SNPs and windows to predict AP in subsequent batches. The pooled simple correlation between genomic prediction values (GPV) and adjusted AP phenotypes was 0.18 in R8 - R11 (n = 412) when 56,424 SNPs were used. When GPV were derived using the most informative SNP from each of the top 10% windows or all SNPs from the top 10% windows identified in training, rGPV,AP was 0.18 and 0.12, respectively. Weaker correlations were obtained when the most informative SNP or all of the SNPs from the top 1% windows were used for prediction (0.01 and 0.06, respectively). These results showed that a limited number of SNPs were able to explain proportions of phenotypic variation similar to that obtained from high-density SNP panels.
Advisor: Daniel Ciobanu