Genomic prediction and genome-wide association study: an application of quantitative genetics in plant breeding programs

Abstract

The development of new tools and advances in high throughput genomic technologies have facilitated genomic selection the identification of sources of variation, especially of complex traits. Therefore, the availability of abundant and cheap markers made it possible to exploit the marker information in breeding programs. The most common tools used in breeding programs that exploit the dense marker coverage are genomic prediction and genome-wide association studies. In the genomic prediction, marker parameters are estimated from a training dataset with genotyped and phenotyped individuals. Subsequently, the trained model is used to predict performance for individuals that are only genotyped. On the other hand, genome-wide association studies test marker-trait associations that may be responsible for the causal variation of interest. We investigated the performance of different genomic prediction models to select parents in the early stage of a hybrid breeding program using estimated general combining ability and their impact on selection accuracy and long-term genetic gain. We evaluated the performance of five genomic prediction models under different SNP marker densities or QTL genotypes using stochastic simulations of an entire hybrid breeding program. We also investigated the ability of univariate and multivariate GWAS identifying markers linked to loci that contribute to resistance to Diplodia ear rot or Fusarium ear rot or both diseases in maize inbred lines. We evaluated the univariate and multivariate approaches using a maize diverse panel evaluated for three different traits.