Incorporating the single-step strategy into a random regression model to enhance genomic prediction of longitudinal traits

Abstract

In prediction of genomic values, the single-step method has been demonstrated to outperform multi-step methods. In statistical analyses of longitudinal traits, the random regression test-day model (RR-TDM) has clear advantages over other models. Our goal in this study was to evaluate the performance of a model that integrates both single-step and RR-TDM prediction methods, called the single-step random regression test-day model (SS RR-TDM), in comparison with the pedigree-based RR-TDM and genomic best linear unbiased prediction (GBLUP) model. We performed extensive simulations to exploit the potential advantages of SS RR-TDM over the other two models under various scenarios with different levels of heritability, number of quantitative trait loci, as well as selection scheme. SS RR-TDM was found to achieve the highest accuracy and unbiasedness under all scenarios, exhibiting robust prediction ability in longitudinal trait analyses. Moreover, SS RR-TDM showed better persistency of accuracy over generations than the GBLUP model. In addition, we also found that the SS RR-TDM had advantages over RR-TDM and GBLUP in terms of its being a real data set of humans contributed by the Genetic Analysis Workshop 18. The findings of our study demonstrated the feasibility and advantages of SS RR-TDM, thus enhancing the strategies for genomic prediction of longitudinal traits in the future.