Building a Calibration Set for Genomic Prediction, Characteristics to Be Considered, and Optimization Approaches

Abstract

The efficiency of genomic selection strongly depends on the prediction accuracy of the genetic merit of candidates. Numerous papers have shown that the composition of the calibration set is a key contributor to prediction accuracy. A poorly defined calibration set can result in low accuracies, whereas an optimized one can considerably increase accuracy compared to random sampling, for a same size. Alternatively, optimizing the calibration set can be a way of decreasing the costs of phenotyping by enabling similar levels of accuracy compared to random sampling but with fewer phenotypic units. We present here the different factors that have to be considered when designing a calibration set, and review the different criteria proposed in the literature. We classified these criteria into two groups: model-free criteria based on relatedness, and criteria derived from the linear mixed model. We introduce criteria targeting specific prediction objectives including the prediction of highly diverse panels, biparental families, or hybrids. We also review different ways of updating the calibration set, and different procedures for optimizing phenotyping experimental designs.

Genomic selection for salinity tolerance in japonica rice

“…The relatedness between the RP and the BP, with a similar structure in the two populations, may also explain this result. Indeed, the genetic distance between the training set and the validation set has been shown to be one of the major factors affecting accuracy (Rio et al, 2022;Scutari et al, 2016). BP41 constitutes only a small subset of the entire BP, but most of the parental accessions and closely related lines, were present in the RP used to train the model.…”

Sparse multi‐trait genomic prediction under balanced incomplete block design

“…Approach 2 consists of guaranteeing high‐quality phenotypic and genotypic data to obtain high‐quality outputs. Approach 3 focuses on optimizing the training and testing set to guarantee predictions in the testing set according to the complexity of the traits; many publications corroborate that the composition of the training sets is essential in improving prediction accuracy (Crossa et al., 2017; Rio et al., 2022). Under this third approach, many strategies for selecting a training set given a testing set have been proposed, while many other strategies for creating a testing set given a training set have also been studied and evaluated (Lopez‐Cruz et al., 2020; Lopez‐Cruz & de los Campos, 2021)…”

“…Approach 2 consists of guaranteeing highquality phenotypic and genotypic data to obtain high-quality outputs. Approach 3 focuses on optimizing the training and testing set to guarantee predictions in the testing set according to the complexity of the traits; many publications corroborate that the composition of the training sets is essential in improving prediction accuracy (Crossa et al, 2017;Rio et al, 2022). Under this third approach, many strategies for selecting a training set given a testing set have been proposed, while many other strategies for creating a testing set given a training set have also been studied and evaluated (Lopez-Cruz et al, 2020;Lopez-Cruz & de los Campos, 2021) Under the third approach, when the genotypes are evaluated in multi-environmental trials (METs), candidate genotypes are evaluated under different environmental conditions, and breeders can select stable genotypes across environments and in specific environments since the genotype × environment (GE) interaction is modeled.…”

Portability of genomic predictions trained on sparse factorial designs across two maize silage breeding cycles

“…Although a simulation work validated the advantage of factorial compared to tester TRSs to predict hybrid values across breeding cycles (Seye et al 2020), further experimental validation is needed. From one cycle to the next, the average relatedness between the TRS and PS decreases and the joint effect of selection, drift, and recombination events change allele frequencies and the linkage disequilibrium between markers and QTLs, which decrease prediction accuracy if the TRS is not updated along cycles (Pszczola et al 2012;Isidro y Sánchez and Akdemir 2021;Rio et al 2022b). This raises questions about how to efficiently update the TRS to maximize prediction accuracy while minimizing phenotyping costs.…”

“…Different optimization criteria have been proposed to define the TRS (Rio et al 2022b). Rincent et al (2012) proposed optimizing the TRS by maximizing the mean of the coefficient of determination (CDmean) of contrasts between each unphenotyped PS individual and the target population mean.…”