Genomic prediction with whole-genome sequence data in intensely selected pig lines

Abstract

Background: Early simulations indicated that whole-genome sequence data (WGS) could improve prediction accuracy and its persistence across generations and breeds. However, results in real datasets have been ambiguous so far. Large data sets that capture most of the genome diversity in a population must be assembled so that allele substitution effects are estimated with higher accuracy. The objectives of this study were to use a large pig dataset to assess the benefits of using WGS for genomic prediction compared to using commercial marker arrays, to identify scenarios in which WGS provides the largest advantage, and to identify potential pitfalls for its effective implementation. Methods: We sequenced 6,931 individuals from seven commercial pig lines with different numerical size. Genotypes of 32.8 million variants were imputed for 396,100 individuals (17,224 to 104,661 per line). We used BayesR to perform genomic prediction for 8 real traits and 9 simulated traits with different genetic architectures. Genomic predictions were performed using either data from a marker array or variants preselected from WGS based on linkage disequilibrium, functional annotation, or association tests. Both single and multi-line training sets were explored. Results: Using WGS improved prediction accuracy relative to the marker array, provided that training sets were sufficiently large, especially for traits with high heritability and low number of quantitative trait nucleotides. The performance of each set of predictor variants was not robust across traits and lines. The most robust results were obtained when preselected variants with statistically significant associations were added to the marker array. Under this method, average improvements of prediction accuracy of 2.5 and 4.2 percentage points were observed in within-line and multi-line scenarios, respectively, with training sets of around 80k individuals. Conclusions: Our results evidenced the potential for WGS to improve genomic prediction accuracy in intensely selected pig lines. Although the prediction accuracy improvements achieved so far were modest at best, we would expect that more robust improvements could be attained with a combination of larger training sets and optimised pipelines.