When less can be better: How can we make genomic selection more cost-effective and accurate in barley?

Abstract

We were able to obtain good prediction accuracy in genomic selection with ~ 2000 GBS-derived SNPs. SNPs in genic regions did not improve prediction accuracy compared to SNPs in intergenic regions. Since genotyping can represent an important cost in genomic selection, it is important to minimize it without compromising the accuracy of predictions. The objectives of the present study were to explore how a decrease in the unit cost of genotyping impacted: (1) the number of single nucleotide polymorphism (SNP) markers; (2) the accuracy of the resulting genotypic data; (3) the extent of coverage on both physical and genetic maps; and (4) the prediction accuracy (PA) for six important traits in barley. Variations on the genotyping by sequencing protocol were used to generate 16 SNP sets ranging from ~ 500 to ~ 35,000 SNPs. The accuracy of SNP genotypes fluctuated between 95 and 99%. Marker distribution on the physical map was highly skewed toward the terminal regions, whereas a fairly uniform coverage of the genetic map was achieved with all but the smallest set of SNPs. We estimated the PA using three statistical models capturing (or not) the epistatic effect; the one modeling both additivity and epistasis was selected as the best model. The PA obtained with the different SNP sets was measured and found to remain stable, except with the smallest set, where a significant decrease was observed. Finally, we examined if the localization of SNP loci (genic vs. intergenic) affected the PA. No gain in PA was observed using SNPs located in genic regions. In summary, we found that there is considerable scope for decreasing the cost of genotyping in barley (to capture ~ 2000 SNPs) without loss of PA.

A connected half‐sib family training population for genomic prediction in barley

“…Linkage disequilibrium estimates in this population were extensive (average LD decay below .2 r 2 : 143 Mb). Several GP studies in barley have reported stable PA as marker density decreases (Abed et al, 2018;Lorenz et al, 2012;Nielsen et al, 2016), which support similar results in other crops (Lorenzana& Bernardo, 2009;Heffner et al, 2011;Spindel et al, 2016). Our results similarly support previous findings that marker density can be substantially reduced without a decrease in PA in malting barley.…”

“…Prediction accuracies from across-family prediction in CV A1 were high and resembled previously published results for heading date (.55) (Abed et al, 2018), plant height (.36-.86) (Lorenz & Smith, 2015;Lorenzana & Bernardo, 2009), and grain protein (.4-.82) (Lorenzana & Bernardo, 2009;Nielsen et al, 2016). Lorenzana and Bernardo (2009) reported PA in biparental families for height (.90) and grain protein (.83) that surpass results from CV A1 and CV W1 in this study, but their datasets include 9-16 environments.…”

“…Studies incorporating GBS markers in genomic prediction have been published in wheat (Triticum aestivum L.) Rutkoski et al, 2014), but most barley genomic prediction studies have used genotyping arrays (Nielsen et al, 2016;Sallam et al, 2015;Schmidt et al, 2016). Abed et al (2018) found that GBS marker density in an advanced barley breeding population could be reduced from ∼30,000 markers to 2000 markers without loss of PA but also noted a large tradeoff between high multiplexing and missing data (2018). The TP described here Crop Science F I G U R E 4 Average cross-validated prediction accuracy for each trait in within-family cross-validation scenario CV W4 .…”

“…Classic NAMs are very large, time consuming to create and phenotype, and often include highly diverse, unadapted founders. Founder lines in a NAM are densely genotyped or fully sequenced to provide dense genome-wide Crop Science markers for association mapping, but high PA in genomic prediction can be obtained with low marker density (Abed, Pérez-Rodríguez, Crossa, & Belzile, 2018;Lorenz et al, 2012;Lorenzana & Bernardo, 2009).…”

NanoGBS: A Miniaturized Procedure for GBS Library Preparation

“…In addition, recent studies have documented that relatively low-density genotyping can be as efficient as high-density genotyping in many cases (Raoul et al, 2016). For example, Abed et al (2018) showed that no significant decrease in the accuracy of genomic selection was seen when using as few as~1K SNPs compared to using 35K SNPs, even in a species such as barley that has a large genome (5.3 Gb). Gorjanc et al (2015) explored and evaluated the impact of increased multiplexing level by decreasing sequencing coverage (from 20 to 0.05x) in livestock genomic selection populations of 500k to 50k individuals.…”