Inclusion of covariables in genome‑wide selection models for prediction accuracy

Abstract

The objective of this work was to evaluate models using the significant single nucleotide polymorphisms (SNPs) detected by marker-assisted selection and genome-wide association, as a fixed effect in the models commonly used in genome-wide selection for F₂ population, in comparison with models using all SNPs. For all models, the Bayesian ridge regression method was used. Comparisons between the models were carried out to evaluate the phenotypic and genotypic prediction ability, phenotypic accuracy, selection gain, coincidence index, and processing time. Both methods failed to accurately identify true quantitative trait loci (QTL). The selection based only in the QTL identified by the studied methods elected individuals of low genetic value. The use of a genome-wide selection model – with the significant SNPs found by the genome-wide association as a fixed effect, and the remaining SNPs as a random effect – was the suitable strategy to select superior individuals with high accuracy. The introduction of QTL already described for a given trait into the genome-wide selection model allows of the selection of superior individuals with greater precision.