Off-campus South Dakota State University users: To download campus access theses, please use the following link to log into our proxy server with your South Dakota State University ID and password.
Non-South Dakota State University users: Please talk to your librarian about requesting this thesis through interlibrary loan.
Thesis - University Access Only
Master of Science (MS)
Since genomic selection was introduced, there have been many statistical methods proposed to calculate accurate genomic enhanced EBV (GE-EBV) for a population of animals. However, accuracies of these methods are inconsistent and do not transpose across populations. One possible reason for this inconsistency is the presence of discovery bias as a consequence of weighting within genomic evaluations. The objective of this project was to create a new statistical method with which to calculate GE-EBV that sufficiently accounts for discovery bias, returning accuracies closer to expected true accuracy of the evaluation. Breeding values were calculated for 2,600 terminal progeny grouped into 107 sire groups first for a full model and then for a model removing one sire group at a time. True breeding values for each model were simulated from 35 moderate frequency (q = 0.29- 0.30) QTN. Marker effects of those QTN were simulated following three different models, constant (CON), randomized-half constant (HALF), and double exponential (DE), with DE being extensively reported. These effects were then used to calculate EBV and corrected EBV (CEBV) of the animals using 2,486 SNP. Results across simulations show that removing sire groups and estimating their breeding values decreases the realized accuracy of CEBV (r = 0.620 ± 0.007) past that of the correlation of TBV with phenotype (r = 0.668 ± 0.008). The overestimated model derived accuracies of both viii CEBV (r = 0.954 ± 7.59e-5) and EBV (r = 0.960 ± 7.11e-5) suggest the susceptibility of similar models to discovery bias. However, the model derived accuracy of CEBV did not decrease by enough to successfully account for discovery bias. Once EBV and CEBV were estimated, they were incorporated into a two trait model separately, along with phenotype, to estimate genomically enhanced estimated breeding values (GE-EBV). Realized accuracies for both GE-EBV and GE-CEBV increased while model derived accuracies only decreased for GE-CEBV. The two trait analysis of CEBV proved to be a successful method to partially account for discovery bias in genomic evaluations where markers are weighted. However, future research is necessary to establish methodology that will fully account for discovery bias in genomic selection.
Library of Congress Subject Headings
References on pages 88-99
Number of Pages
South Dakota State University
In Copyright - Educational Use Permitted
Parham, Jamie Tart, "Partially Accounting for Discovery Bias in Genomic Selection by Conditioning Out Subsets of the Population During Training" (2015). Electronic Theses and Dissertations. 1780.