Accuracy of non-additive models and population structure of genomic selection in maize

Abstract

Genomic selection (GS) is a method to predict the genetic value of selection candidates based on the genomic estimated breeding value (GEBV) predicted from high-density markers positioned throughout the genome. The accuracy of prediction of hybrid performance by GS has been examined in maize. However, there are many challenges in GS to increase the accuracy of the estimation of GEBV. For example, the genetic relatedness between individuals, the non-additive effects and the population structure in the training and prediction populations affects accuracy and the gains per selection. Thus, the aim of this work is to evaluate the accuracy and predictive ability of non-additive models of GS in the presence of different levels of population structure in maize. The training population of maize will be simulated to have 1176 F1 individuals phenotyped for grain yield. A real genotypic data of forty-nine lines (parents) will be used to obtain the genotypic data of the simulated hybrids. The genotyping of these lines was performed in the Affymetrix® platform by using an array with, approximately, 660,000 SNPs for maize. In possession of phenotypic and genotypic data, will be performed the evaluation of the models additive, additive-dominant and additive-dominant-epistatic of GS. The model-based approach implemented in software package STRUCTURE v2.3.3 will be used to reveal population structure of the 49 maize lines. The additive, additive-dominant and additive-dominant-epistatic models will be compared through the accuracy and predictive capacity observed in each of the models used at different levels of population structure. The statistical genetic analysis of GS will be implemented by the software R (3.1.1) by the packages rrBLUP and ASreml-R. (AU)