2013 Annual Report
1a.Objectives (from AD-416):
As described in grant entitled “Development and field evaluation of genome-wide marker-assisted selection (GWMAS) over multiple generations in commercial poultry,” the consortium that includes members from Wageningen University will:
1. Refine the Theoretical and Molecular Aspects of GWMAS; and
2. Field Assessment of GWMAS
1b.Approach (from AD-416):
For Objective #1, alternative methods of utilizing GWMAS will also be explored. Principal Component Regression (PCR) and Partial Least Squares (PLS) are two methods that were mainly developed in chemometrics to deal with situations where the number of effects to be estimated greatly exceeds the number of records. The estimation of 10-100,000 haplotype effects from ~1,000 phenotypic records is exactly such a situation, and thus may suit these methods very well. Semi-parametric regression was suggested by Gianola et al. (2006) for the estimation of GW-EBVs. Its main advantages are: (1) it ‘automatically’ deals with non-additivity of gene effects, whereas the other methods, in their basic form, assume additive gene action; and (2) it makes few distributional assumptions about the data, which could make this method more robust in practical situations.
A computer simulation study based on data generated by Muir’s (2007) program will be conducted to compare the statistical methods Ridge Regression (RR), BLUP, GRM, BayesB, Xu’s (2003), PCR, PLS, and semi-parametric regression for their computational efficiency and accuracy of predicting breeding values in poultry breeding programs, their bias (over/under prediction) of true breeding values, their sensitivity to the assumptions that are made about the genetic model, and the number of records and genotypes that are computationally feasible. The simulated genome structure will largely follow that of Meuwissen et al. (2001), but the parameters determining the genome structure, such as effective population size, mutation rates and distribution of mutational effects, gene action (additivity, dominance, epistatic effects), admixture of populations, etc., will be varied, in order to assess the sensitivity of the methods to differences in the structure of the genome.
For Objective #2, the expected breeding values (EBV) will be computed for each normalized trait of each animal based on genotypic information and phenotypes (if present). These EBV would in turn be used to determine total merit by use of the appropriate weights given by the companies. The companies will do their own BLUP evaluations based on phenotype as per their usual breeding programs. The goal is to make selection decisions based on the EBVs for each trait and/or total merit the same for each method of selection [BLUP, GWMAS], the only difference being the way in which the EBVs will be estimated.
This project is directly linked to project 3635-31320-009-05R titled "Development and Field Evaluation of Genome-Wide Marker-Assisted Selection (GWMAS) Over Multiple Generations in Commercial Poultry." The advantages of genomic selection (GWMAS) are well understood. In summary, GWMAS is most advantageous for traits that are difficult to or costly to measure (carcass quality, feed efficiency) or are collected late in life (full record egg production). This technology is the final realization of genomics, i.e., direct selection on the genotype and can double response to selection per unit of time for some traits. However, we have shown using simulations that a loss in accuracy will result once selection on those markers has commenced. Among the unanswered questions are.
1)how rapidly the decline in accuracy will occur and.
2)how often the retraining process must occur and how much of the lost accuracy can be regained? To address these questions, we analyzed three traits from two pure lines of broiler type chickens across 5 generations. The complete populations consisted of 183,784 and 164,246 broilers for two lines. The genotyped subsets consisted of 3,284 and 3,098 broilers with 57,636 SNPs. The training population consisted of the first two generations. Following 3 generations of selection based on ssGBLUP (genomic selection), we compared the accuracy with which EBVs could be predicted in those 3 generations based only on the genotypes (ssGBLUP) or pedigree (BLUP). Accuracy was determined as the correlation between predicted breeding value and the phenotype divided by the square root of heritability for that trait. Results showed that the relative accuracy for ssGBLUP remained about 33% higher than BLUP in all generations across both lines. The results differ from the predictions of the simulations from Muir (2007) who used the same set of assumptions as Meuwissen et al (2001). Our conclusion is that the results are characteristic of the infinitesimal model with thousands of quantitative trait locus (QTL) rather than a 100 or fewer as assumed by Meuwissen et al. (2001). There are a number of important conclusions from our results. The infinitesimal model is the most likely the correct approximation to the true biological genetic variation for these traits which implies:.
1)ssGBLUP is more accurate than any SNP based method (e.g., BayesA, BayesB, or other such methods, which assume a relatively small number of SNPs whose effects can be estimated and selected for; in contrast, the infinitesimal model implies that genomic selection improves accuracy by increasing the precision with which relationships are estimated),.
2)The results also imply that GWAS for these traits will find relatively few large effect QTLs,.
3)Missing heritability type issues will result if selection is based on small chip with large effects QTL, i.e. will account for small amount of total genetic variation,.
4)The results further imply that for these traits it will be possible to do genotype only selection for several generations before retraining will be necessary. In general these results are very favorable for genomic selection suggesting that such approaches will be useful in commercial breeding programs.