Dimensionality of genomic information and its impact on genome-wide associations and variant selection for genomic prediction: a simulation study

Abstract

Background Identifying true positive variants in genome-wide associations (GWA) depends on several factors, including the number of genotyped individuals. The limited dimensionality of genomic information may give insights into the optimal number of individuals to be used in GWA. This study investigated different discovery set sizes based on the number of largest eigenvalues explaining a certain proportion of variance in the genomic relationship matrix (G). In addition, we investigated the impact on the prediction accuracy by adding variants, which were selected based on different set sizes, to the regular single nucleotide polymorphism (SNP) chips used for genomic prediction. Methods We simulated sequence data that included 500k SNPs with 200 or 2000 quantitative trait nucleotides (QTN). A regular 50k panel included one in every ten simulated SNPs. Effective population size (Ne) was set to 20 or 200. GWA were performed using a number of genotyped animals equivalent to the number of largest eigenvalues of G (EIG) explaining 50, 60, 70, 80, 90, 95, 98, and 99% of the variance. In addition, the largest discovery set consisted of 30k genotyped animals. Limited or extensive phenotypic information was mimicked by changing the trait heritability. Significant and large-effect size SNPs were added to the 50k panel and used for single-step genomic best linear unbiased prediction (ssGBLUP). Results Using a number of genotyped animals corresponding to at least EIG98 allowed the identification of QTN with the largest effect sizes when Ne was large. Populations with smaller Ne required more than EIG98. Furthermore, including genotyped animals with a higher reliability (i.e., a higher trait heritability) improved the identification of the most informative QTN. Prediction accuracy was highest when the significant or the large-effect SNPs representing twice the number of simulated QTN were added to the 50k panel. Conclusions Accurately identifying causative variants from sequence data depends on the effective population size and, therefore, on the dimensionality of genomic information. This dimensionality can help identify the most suitable sample size for GWA and could be considered for variant selection, especially when resources are restricted. Even when variants are accurately identified, their inclusion in prediction models has limited benefits.

Marker effect p-values for single-step GWAS with the algorithm for proven and young in large genotyped populations

“…The new peaks had clear linkage disequilibrium trails, meaning that ssGWAS resolution increases as more genotyped animals are included in the analyses. As previously shown, especially for populations with a small effective population size ( Ne ) and more polygenic traits, increasing the genotype set reduces the estimation error and the shrinkage of SNP effects, which increases the power of discovering significant variants (Lourenco et al, 2017; Jang et al, 2023; Misztal et al, 2023). The benefit of an increase in the genotype set size can also be observed when comparing Approx_GinvAPY with approx_GinvAPY450K.…”

“…While evaluating two simulated populations with the same Ne, Misztal et al (2023) observed that increasing the number of individuals contributing with genotypes and phenotypes by three times increased the correct identification of significant SNPs. Similarly, Jang et al (2023) showed that for highly polygenic traits (2000 QTN) with Ne of 20 and a moderate heritability of 0.30, no QTN was accurately identified until a complete genotype set, composed of 30K genotyped animals, was included in the analyses. For livestock populations with even smaller Ne and traits of lower heritability, such as reproduction and fitness traits, QTN identification may be even more challenging, especially when limitations exist on the amount of genomic information used in the estimation process.…”

The big challenge for livestock genomics is to make sequence data pay

“…They were unsuccessful in estimating these effects accurately from genome-wide association studies, presumably due to linkage disequilibrium. Jang et al (Jang et al, 2023) simulated the process of pre-selection by genomewide association studies, exploring under which conditions large effects can be identified to supplement the SNP chip. They concluded that:…”

Preselecting Variants from Large-Scale Genome-Wide Association Study Meta-Analyses Increases the Genomic Prediction Accuracy of Growth and Carcass Traits in Large White Pigs

“…In recent years, preselected potential causal variants identified through genome-wide association studies (GWASs) have already been applied in genomic prediction to improve the prediction accuracy for a variety of traits in livestock and poultry [19,20], including pigs [21,22]. In certain instances, this strategy has resulted in improved accuracy of genomic prediction for growth and carcass traits in pigs, with improvements ranging from 0.9 to 46% for multi-breed populations [21][22][23]. However, it should be noted that this strategy did not result in improved prediction accuracy in all cases [21,24,25].…”

“…A simulation study indicated that the detection of causal variants explained only 20% of the total genetic variance when using a sample size of 7000 for a GWAS in an effective population size of 20 [23]. Recently, Ros-Freixedes et al [21] used preselected SNPs based on GWAS results for genomic prediction, studying a population of nearly 100,000 pigs from the largest lines with imputed WGS data.…”