Accounting for epistasis improves genomic prediction of phenotypes with univariate and bivariate models across environments

Elaheh Vojgani,Armin C Hölker,Manfred Mayer,Johannes W R Martini,Torsten Pook,Chris-Carolin Schön,Henner Simianer

doi:10.1007/s00122-021-03868-1

Abstract

Key MessageThe accuracy of genomic prediction of phenotypes can be increased by including the top-ranked pairwise SNP interactions into the prediction model.We compared the predictive ability of various prediction models for a maize dataset derived from 910 doubled haploid lines from two European landraces (Kemater Landmais Gelb and Petkuser Ferdinand Rot), which were tested at six locations in Germany and Spain. The compared models were Genomic Best Linear Unbiased Prediction (GBLUP) as an additive model, Epistatic Random Regression BLUP (ERRBLUP) accounting for all pairwise SNP interactions, and selective Epistatic Random Regression BLUP (sERRBLUP) accounting for a selected subset of pairwise SNP interactions. These models have been compared in both univariate and bivariate statistical settings for predictions within and across environments. Our results indicate that modeling all pairwise SNP interactions into the univariate/bivariate model (ERRBLUP) is not superior in predictive ability to the respective additive model (GBLUP). However, incorporating only a selected subset of interactions with the highest effect variances in univariate/bivariate sERRBLUP can increase predictive ability significantly compared to the univariate/bivariate GBLUP. Overall, bivariate models consistently outperform univariate models in predictive ability. Across all studied traits, locations and landraces, the increase in prediction accuracy from univariate GBLUP to univariate sERRBLUP ranged from 5.9 to 112.4 percent, with an average increase of 47 percent. For bivariate models, the change ranged from −0.3 to + 27.9 percent comparing the bivariate sERRBLUP to the bivariate GBLUP, with an average increase of 11 percent. This considerable increase in predictive ability achieved by sERRBLUP may be of interest for “sparse testing” approaches in which only a subset of the lines/hybrids of interest is observed at each location.

Highlights

Univariate Genomic Best Linear Unbiased Prediction (GBLUP) within the environment is used as a reference and is compared to results obtained with univariate Epistatic Random Regression BLUP (ERRBLUP) within environments and univariate sERRBLUP when the top 5, 1, 0.1, 0.01, 0.001, and 0.0001 percent of pairwise SNP interactions are maintained in the model
In this study, estimated effect variances were identified as the best selection criteria in sERRBLUP, since sERRBLUP predictive abilities were observed to be more robust when the selection of pairwise SNP interaction was based on the effect variances compared to absolute effect sizes, especially when the top 0.001 and 0.0001 percent of interactions are maintained in the model (Fig. S10 and S11)
The maximum predictive ability obtained from univariate sERRBLUP is almost identical when selecting SNP interactions based on absolute effect sizes or effect variances for both KE and PE (Fig. S12)

Summary

Introduction

Genomic prediction of phenotypes has been widely explored for crops (Crossa et al 2010), livestock(Daetwyler et al 2013), and clinical research (de los Campos et al 2013).Communicated by Antonio Augusto Franco Garcia.Broad availability and cost-effective generation of genomic data had a considerable impact on plant (Bernardo and Yu 2007; de los Campos et al 2009; Crossa et al 2010, 2011; de Los Campos et al 2010; Pérez et al 2010) and animal breeding programs (de los Campos et al 2009; Hayes and Goddard 2010; Daetwyler et al 2013). Genomic prediction relates a set of genome-wide markers to the variability in the observed phenotypes and enables the prediction of phenotypes or genetic values of genotyped but unobserved material (Meuwissen et al 2001; Jones 2012; Windhausen et al 2012). This approach has been positively evaluated in most major crop and livestock species (Albrecht et al 2011; Daetwyler et al 2013; Desta and Ortiz 2014) and is becoming a routine tool in commercial and public breeding programs (Stich and Ingheland 2018). In plant breeding, phenotyping is one of the major current bottlenecks and the optimization or minimization of phenotyping costs within breeding programs is needed

Methods

Results

Conclusion