Title: The performance of genotype-to-phenotype models accounting for large-effect loci, epistasis, and pleiotropy
Authors: Alexander Lipka - University of Illinois (United States) [presenting]
Abstract: Models that reflect the multifaceted contributions of genomic loci have a potential to facilitate unprecedented quantification of the genetic architecture underlying various traits and increase genomic selection (GS) prediction accuracies. The performance of statistical models is evaluated for traits with contrasting genetic architectures. These traits were simulated using marker data in maize, sorghum, and humans. These simulation studies revealed that including peak-associated markers from a genome-wide association study (GWAS) of a training set as fixed-effect covariates in an RR-BLUP genomic selection model is capable of decreasing prediction accuracy, increasing the variability of prediction accuracy across replicate traits, and increasing the bias of predictions compared to a standard RR-BLUP GS model. The studies also suggest that a model quantifying the simultaneous contribution of additive and two-way epistatic loci is capable of identifying and distinguishing between simulated additive and epistatic quantitative trait nucleotides (QTNs). Finally, the latest results from an ongoing simulation study seeking to explore the ability of a multi-trait GWAS model to identify simulated pleiotropic QTN is presented. These results will underscore current efforts to refine GS models that that go beyond univariate models accounting for only additive marker effects.