Use of whole-genome sequence data for genomic prediction across populations and species

Abstract

Raymond, B. (2020). Use of whole-genome sequence data for genomic prediction across populations and species. PhD thesis, Wageningen University, the Netherlands The availability of whole genome sequence data presents an opportunity to improve the accuracy of genomic prediction (GP), given the expectation that the data contains the causal variants that underlie any given complex trait. This thesis investigated the benefit of using whole genome sequence data for GP across populations and species. Results show that the accuracy of GP across populations does not improve by simply increasing marker density up to whole genome sequence level (Chapter 2). However, accuracy can be improved if GP across populations is based on a few variants that are pre-selected from whole genome sequence based on the significance level of their effect on the trait from a genomewide association study (GWAS). The result highlights the relevance of GWAS for GP. In Chapter 3, it was demonstrated that the accuracy of prediction can further be improved using a multi-population GP model in which important pre-selected variants are used to create a genomic relationship matrix (GRM), other available and unselected variants are used to create a second GRM, and both GRMs are fitted simultaneously. While the pre-selected variants in the first GRM are isolated from the noise effect of neutral variants, and as such their effects are more accurately estimated, the variants in the second GRM captures genetic variance for the trait that cannot be captured by the pre-selected variants in the first GRM. In terms of accuracy, it was shown that the multi-population, multiple GRMs (MPMG) GP model outperforms within-population and multi-population GP models in which either the pre-selected or all available variants are equally weighted in a single GRM. In Chapter 4, the predictive performance of the MPMG model was theoretically underpinned by deriving and validating a deterministic prediction equation for its accuracy. Using the derived prediction equation, it was found that the predictive performance of the model is due to its ability to benefit from the low number of effective chromosomal segments () represented by the few preselected variants in the first GRM. However, the low values for due to the preselected variants is an advantage for the MPMG model only if variant pre-selection is accurate, such that the pre-selected variants explain some genetic variance for the trait of interest. In addition to its use as a tool to gain theoretical insights into the performance of biology-informed GP models, the derived prediction equation can be used to, a priori, estimate expected accuracy if the MPMG model were to be implemented for GP. In Chapter 5, the usefulness of summary-level GWAS result for human height as prior information for identifying genes and gene-associated variants that affect stature in cattle, was investigated. Results show that in some cases, for example in the absence of stature GWAS, human height GWAS results can be useful in identifying cattle...