Training set optimization under population structure in genomic selection

Abstract

Key messagePopulation structure must be evaluated before optimization of the training set population. Maximizing the phenotypic variance captured by the training set is important for optimal performance.AbstractThe optimization of the training set (TRS) in genomic selection has received much interest in both animal and plant breeding, because it is critical to the accuracy of the prediction models. In this study, five different TRS sampling algorithms, stratified sampling, mean of the coefficient of determination (CDmean), mean of predictor error variance (PEVmean), stratified CDmean (StratCDmean) and random sampling, were evaluated for prediction accuracy in the presence of different levels of population structure. In the presence of population structure, the most phenotypic variation captured by a sampling method in the TRS is desirable. The wheat dataset showed mild population structure, and CDmean and stratified CDmean methods showed the highest accuracies for all the traits except for test weight and heading date. The rice dataset had strong population structure and the approach based on stratified sampling showed the highest accuracies for all traits. In general, CDmean minimized the relationship between genotypes in the TRS, maximizing the relationship between TRS and the test set. This makes it suitable as an optimization criterion for long-term selection. Our results indicated that the best selection criterion used to optimize the TRS seems to depend on the interaction of trait architecture and population structure.Electronic supplementary materialThe online version of this article (doi:10.1007/s00122-014-2418-4) contains supplementary material, which is available to authorized users.

Updating the reference population to achieve constant genomic prediction reliability across generations

“…The slopes in Reg1 were expected to be positive, as higher relationships between RP and EVA lead a higher reliability (Habier et al, 2010;Clark et al, 2012;Pszczola et al, 2012a;Rincent et al, 2012;Isidro et al, 2015;Wu et al, 2015). This was confirmed by the results of the current study as shown in Table 1.…”

“…This was confirmed by the results of the current study as shown in Table 1. In addition, low relationships in RP are desired (Pszczola et al, 2012a;Rincent et al, 2012;Isidro et al, 2015;Wu et al, 2015). Therefore, the slope in Reg2, which uses the off-diagonal elements of the inverted G, is expected to be positive (as confirmed by the results in Table 2).…”

“…Such procedure would, however, require a complex optimization algorithm which on one hand would maximize the relationships between RP and EVA but on the other hand would minimize the relationship of the update with already existing RP. Other studies showed that random sampling of the RP may be suboptimal and attempted to derive algorithms based on statistical properties of the genomic prediction model (Rincent et al, 2012;Isidro et al, 2015). Although these approaches may be theoretically more optimal, the advantage of our approach based on relationships is its' intuitiveness leading to easier application in practice.…”

“…Another important aspect is the RP design. An optimally designed RP is expected to maximize the reliability for a given population (Pszczola et al, 2012a;Rincent et al, 2012;Isidro et al, 2015). Several studies have shown that across scenarios the highest accuracies were obtained when the RP consisted of animals that are distantly related to each other (Clark et al, 2012;Pszczola et al, 2012a;Rincent et al, 2012;Isidro et al, 2015;Wu et al, 2015) and closely related to the evaluated animals (EVA) (Habier et al, 2010;Pszczola et al, 2012a;Wu et al, 2015).…”

Accelerating genetic gains in legumes for the development of prosperous smallholder agriculture: integrating genomics, phenotyping, systems modelling and agronomy

“…The genetic relatedness or population structure (Saatchi et al, 2011;Riedelsheimer et al, 2013;Wray et al, 2013) may result in overestimating the heritability of the traits (Price et al, 2010;Visscher et al, 2012;Wray et al, 2013). The population structure of the training population can be determined with greater accuracy using genome wide SNPs compared with the simple sequence repeats and SNP arrays (Isidro et al, 2015). Accuracy of estimating breeding values is important in GS.…”