Single nucleotide polymorphism genotypes for 1,041 exome-sequenced bread wheat accessions were used to select content for the Infinium Wheat Barley 40K SNP array. The accessions included 790 previously reported by He et al. (2019) to capture global wheat (T. aestivum) diversity, an additional 149 accessions selected from the global collection contained in the associated VCF file³ to expand the diversity captured and 102 historical breeding lines from the InterGrain commercial wheat breeding program⁴. The first two sets of accessions maximally captured genetic diversity among 6,087 globally diverse wheat accessions comprising landraces, varieties, synthetic derivatives, and novel trait donor lines (He et al., 2019). The additional 149 accessions were selected to capture genetic diversity within synthetic derivative germplasm derived from crossing 100 primary synthetics (derived from interspecific hybridization of durum wheat with Ae. tauschii) to three Australian varieties: Yitpi, Annuello, and Correll (Ogbonnaya et al., 2007). The latter two sets of accessions were exome-capture sequenced as described in He et al. (2019). SNP discovery was performed using the first two sets of accessions and the resulting SNP list was used to call SNP genotypes across all accessions.

The Infinium 90K wheat SNP genotypes reported by Maccaferri et al. (2019) for a globally diverse tetraploid wheat collection of 1,856 accessions comprising wild emmer (T. turgidum ssp. dicoccoides), domesticated emmer (T. turgidum ssp. dicoccocum), and T. turgidum genotypes including durum landraces and cultivars were used to select tetraploid wheat specific SNP.

A georeferenced landrace collection of 267 exome-sequenced barley accessions, including 2- and 6-rowed H. vulgare landraces as well as H. spontaneum (Russell et al., 2016), and 117 whole genome sequenced accessions representing historical breeding lines from the InterGrain commercial barley breeding program were used to select the content for the SNP array.

In wheat, SNP discovery and genotype calling were performed as described by He et al. (2019), against IWGSC RefSeq v1.0 (The International Wheat Genome Sequencing Consortium [IWGSC], 2018). After filtering for > 60% call rate and > 1% MAF, 2.04M SNPs were used for LD analysis. To filter for nucleotide variation originating from Ae. tauschii, D-genome-specific SNP that had a MAF > 0.1 in the synthetic derivative wheat and MAF < 0.1 in the globally diverse wheat collection were identified. In addition, the top 2% of D-genome SNPs that showed differential allele frequencies between these two groups based on F_st values (Weir and Cockerham, 1984) were selected. From these two SNP sets, SNP uniformly distributed across the D-genome were selected for inclusion as SNP content.

In barley, SNP discovery was performed as described by He et al. (2019) using the exome sequence data published by Russell et al. (2016), against Morex v1.0 (Mascher et al., 2017). Following the removal of H. spontaneum-like accessions based on principal component analysis (PCA) clustering (which left 157 H. vulgare-like accessions), the resulting SNP list was used to call SNP genotypes in the 120 InterGrain historical breeding lines. After filtering for > 60% call rate and > 5% MAF (a higher cut-off was used in barley due to the smaller reference population), 932,098 SNPs were used for LD analysis.

Linkage disequilibrium analysis for the filtered SNP was performed using PLINK (Purcell et al., 2007) at the chromosome level within each species with a maximum window size of 2 Mb; i.e., all the SNPs in a tag SNP set (see “Results” for definition) had to be within a 2 Mb window. The squared correlation coefficient (r²) based on the allele frequency in the global barley or wheat diversity panel (excluding the synthetic derivatives) between two SNPs was considered as a measure of LD.

To maximize the number of SNPs assayed for a given number of probes on the bead chip array, A/T and C/G variants (Infinium Type I SNP which require two probes) were avoided. To maximize SNP scorability and genotype calling accuracy, polymorphism underlying the 50-mer oligonucleotide SNP probe sequences were also avoided as they are known to cause shifts in SNP cluster position (Wang et al., 2014). For tagging SNPs (tSNPs), the probe sequences were required to align uniquely to the target genome and not aligned to the other genome; i.e., a wheat SNP probe had to align uniquely to the wheat genome and not to the barley genome, and vice versa. Finally, an Illumina Design Tool score of ≥ 0.6 was required for a probe to be included as array content. A relaxed set of criteria was also used (to tag SNP sets otherwise missed) which allowed up to three alignments to the target genome.

A custom algorithm was used to select tSNP tagging LD blocks in each of the global collections and to facilitate imputation from the density of the SNP array. In brief, for each chromosome, the algorithm iteratively selected the most informative tSNPs passing all filters (based on its r² value from the LD analysis), removed all SNPs linked to the selected tSNPs from the remaining list of SNPs, as well as all SNPs linked to any SNP in the selected tSNP set to avoid directly tagging any SNP at r² ≥ 0.9 more than once, before repeating the process until a target number of tSNP was reached. This process ensured that the set of tSNP selected was the minimum set required to tag the most SNPs at r² ≥ 0.90. Specifically, for a given set of SNPs S = {s₁,s₂,…} and function r²(s_i,s_j) defining the Pearson correlation coefficient squared ∀s_i,s_j ∈ S, we define the tSNP set for s_i at q to be:

Rename the T^q_{s_i} and define Ts⁢o⁢r⁢t⁢e⁢dq=(Tsjq)j=1n=Ts1q,Ts2q,Ts3q,… where i≥j=>|Tsiq|≤|Tsjq|.

In other words, T^q_sorted is an ordering of equivalent SNP sets, monotonically decreasing in size.

Let F⊂S be a subset of filtered SNPs. Define F⁢(Ts⁢o⁢r⁢t⁢e⁢dq)={Tsjq|sj∈F}.

We define Ts⁢o⁢r⁢t⁢e⁢dq-Tsiq={(Tsjq)sj∈S|Tsiq∩Tsjq=∅}, head(L) to be the first element of the ordered sequence L, and select (Tsiq) = s_i.

The algorithm is then:

For example, the above applied with q = 0.9,m = 10 defines the first iteration of tSNP selection.

To guard against possible loss of imputation accuracy due to SNP assays failing to provide reliable genotype calls, a level of redundancy was included in the tSNP sets for wheat and barley. Specifically, three tSNPs were chosen when the number of SNPs tagged was ≥ 50 and two tSNPs were selected when the number of SNPs tagged was ≥ 20. Single tSNP were selected when they tagged at least 10 SNPs. Some SNP sets could not be tagged because no probe passed all the filters; in this case, we ran the algorithm on the remaining sets allowing for SNP passing relaxed filters (up to three hits to the target genome were allowed). In addition, tSNPs were selected to tag genomic regions that had sparse SNP coverage but high LD; i.e., tagging < 10 SNP within windows larger than 500 Kb in wheat and 1Mb in barley. Finally, SNPs were selected in regions still lacking SNPs after the previous steps.

To ensure broader applicability of the SNP array in research and breeding, the content included SNP selected to specifically interlink germplasm resources, such as the 19,778 domesticated barley accessions with GBS genotypes described by Milner et al. (2019). It also included SNP probes designed to interrogate published trait-linked markers in wheat and barley. Designs for these markers were based directly on published sequences or from the alignment of published primers or flanking sequences and inference of the targeted nucleotide variation. For all trait-linked markers, the best probe design was selected based solely on the Illumina quality score. Due to the difficulty of designing SNP probes targeting known alleles of phenology genes, we selected 293 exome SNPs around the genes reported by Shi et al. (2019).

The wheat and barley global diversity sets were used as reference haplotypes for imputation. For wheat, accessions clustering with the synthetic derivatives in a PCA analysis were excluded. For barley, only samples with < 20% missing data were used. In both species, missing data were filled in using Beagle 4.1 (Browning and Browning, 2007) and phased with Eagle 2.4.1 (Loh et al., 2016). In total, 868 and 155 wheat and barley lines were used as reference haplotypes.

In wheat, SNP coordinates were converted to IWGSC v2.0 pseudomolecules⁵ (Zhu et al., 2021) before imputation. After transfer into the v2.0 assembly, there were 18,521 tSNPs before imputation, with 630,058, 549,003, and 352,947 tagged at r² ≥ 0.50, 0.70, and 0.90, respectively.

To assess the accuracy of imputation into globally diverse germplasm, 100-fold cross validation was performed. A random subset of 100 wheat (or 10 barley) lines had their true genotypes masked, leaving only the tSNP. The remaining lines were then used as the reference population with Minimac3 software (Das et al., 2016) to impute back the missing genotypes for three different target SNP sets: the set of SNPs tagged at r² ≥ 0.50, 0.70, and 0.90. The imputation accuracy for each line, measured as both correlation squared and concordance between the actual and imputed genotypes, was calculated from 100 repetitions of this process in each of wheat and barley. The correlation squared metric used was the Pearson correlation coefficient squared (r²) between SNPs called in both genotypes being compared, while concordance was measured as the fraction of SNPs in agreement with those called in both genotypes being compared.

Genome-wide association studies were performed using the GCTA software (Yang et al., 2011) using a mixed linear model with the SNP matrix fitted as a fixed effect and genomic relationship matrix (GRM) as a random effect. The GRM is a covariance matrix from the SNP information for each sample. Phenotype data for awned status (scored as a presence-absence trait) in 355 wheat accessions and row type (scored as two- or six-rowed) in 121 barley accessions were used. The number of SNPs used in the wheat GWAS, after transfer into the IWGSC v2.0 assembly (see imputation section above), were 18,515 (selected tSNP), 548,864 (imputed tSNP) and 1,086,408 (exome). The number of SNPs used in the barley GWAS were 13,518 (selected tSNP), 359,752 (imputed tSNP), and 1,719,837 (exome). An arbitrary threshold P-value of 1×10^–5 was used as the significant threshold for declaring a marker-trait association.

Samples were assayed following the protocol for Infinium XT bead chip technology (Illumina Ltd., CA, United States). SNP clustering and allele calling was performed using GenomeStudio Polyploid software (Illumina Ltd., CA, United States) using the Illumina-supplied wheat or barley SNP manifest file. The custom genotype calling pipeline described by Maccaferri et al. (2019) was also used.

Figures and plots were produced in R 3.6.1⁶ and ggplot2 (Wickham, 2016). For PCA plots, SNPRelate 1.20.1 (Zheng et al., 2012) was used.

The central idea of the design concept is to exploit LD using the r² measure to define sets of SNPs that can be considered equivalent; for a given SNP (referred to as a tSNP), we define its tag SNP (or tSNP) set as the set of SNPs tagged by this SNP at r² ≥ 0.9. This metric provides a measure of equivalence as well as a natural ranking of SNP by their informativeness, as defined by the size of their tSNP set. We assume that the relationship is symmetrical; i.e., if SNP A is in the tSNP set of SNP B, then SNP B should be in the tSNP set of SNP A. The original set of SNPs are then filtered using technology and application-specific criteria (see section “Materials and Methods”) while maintaining connectivity to SNPs that fail the filters via the tSNP that pass the filters.

To design a genotyping array that has broader applicability in research and breeding, the SNPs should be discovered in diverse germplasm to avoid ascertainment bias (since LD is population-dependent) and with sufficient density to produce large tSNP sets. The latter helps ensure that at least one SNP in a tSNP set will pass all the design filters in most instances. Here, we used a globally diverse set of barley landrace accessions and a globally diverse set of wheat accessions that included landraces, varieties, novel trait donors, and historical breeding lines (Figure 1). For array designs that are focused only on breeding applications, SNP discovery should aim to capture the genetic diversity within the breeding germplasm pool.

A novel selection algorithm (described in “Materials and Methods”) is then used to select SNPs that maximize LD capture, while minimizing the number of SNPs assayed on the array, using only SNPs that pass the design filters.

The design concept can be applied to any animal or plant species. In addition to this set of SNPs, utility in research and breeding can be further enhanced by including context-relevant SNPs, such as trait-linked markers and markers that link germplasm resources across different genotyping technologies. The approach used to design the Wheat Barley 40K SNP array is summarized in Supplementary Figure 5.

Filtering for a minimum minor allele frequency (MAF) of 1% and maximum missing rate of 40% using the 8,869,370 wheat SNP published in He et al. (2019) resulted in 2,037,434 high quality SNPs for downstream analysis. Of these, 122,799 SNPs had at least one array probe that passed all design filters. In barley, filtering of the 1,843,823 SNPs identified from our processing of exome capture sequence from the accessions from Russell et al. (2016) for MAF > 5% and missing rate < 40% resulted in 932,098 high quality SNPs for downstream analysis, of which 119,633 SNPs had at least one array probe passing all the filters. The filtered SNP matrices used in the subsequent analysis are available at https://dataverse.harvard.edu/dataverse/WheatBarley40k_v1.

Based on LD values of r² ≥ 0.9, a total of 1.07M wheat and 413,508 barley high quality SNPs were singletons; i.e., they had no SNP within 1Mb up and downstream with r² ≥ 0.9. These SNPs were either genuine singletons or categorized as singletons due to the absence of additional SNPs within the surrounding 2Mb region. As singleton SNPs can only be tagged directly, which is not feasible on a low-density array, these SNPs were not considered further for inclusion on the array.

The custom selection algorithm grouped the 122,799 non-singleton wheat SNPs passing all design filters into 11,076 tSNP tagging SNP sets containing ≥ 10 SNP within a 2 Mb window. These tSNPs tagged 317,599, 538,326, and 652,476 SNPs at r² ≥ 0.9, 0.7 and 0.5, respectively. Of the 119,633 non-singleton barley SNPs passing all filters, the selection algorithm identified 7,316 tSNPs which tagged a total of 150,096, 294,659, and 390,844 SNPs at r² ≥ 0.9, 0.7, and 0.5, respectively. At the genome level, the rate of return per tSNP was surprisingly similar for wheat and barley and plateaued at about 15,000 tSNP at r² ≥ 0.9 (Figure 2). However, the rate of return per tSNP varied at the chromosome level (Supplementary Figure 1).

In total, 21,012 wheat and 13,469 barley tSNPs were included as content on the array. This tally includes redundant SNPs selected to guard against the possible loss of imputation accuracy due to SNP assays that might fail; SNP passing a relaxed set of filters (allowing up to three alignments to the target genome) and tagging SNP sets untaggable with the strictly filtered SNP; and SNP to tag genomic regions that had sparse SNP coverage but high LD; i.e., tagging < 10 SNPs within windows larger than 500 Kb in wheat and 1 Mb in barley. The latter SNPs are expected to support increased imputation density in these regions as higher density SNP datasets become available in the future. The wheat tSNPs tagged a total of 394,034, 636,641, and 758,452 SNPs at r² ≥ 0.9, 0.70 and 0.50, respectively, while the barley tSNP tagged a total of 187,412, 361,012, and 471,645 SNPs, respectively. Importantly, the MAF distributions for the tSNP, tagged SNP, and filtered SNP from the globally diverse wheat and barley collections closely matched one another (Figure 3). The distribution of the selected tSNP in the wheat and barley genomes is shown in Supplementary Figure 4.

Cross validation (100-fold) was used to assess the accuracy for imputing from the tSNPs on the array to the sets of SNPs tagged at r² ≥ 0.50, 0.70, and 0.90, in the globally diverse wheat and barley germplasm. This was achieved by randomly selecting 100 wheat (or 10 barley) lines and masking their true genotypes to leave only the tSNPs. Using the remaining lines as the reference population, the masked genotypes for each randomly selected line were imputed to the density of one of the target SNP sets. Accuracy was determined from the correlation squared and concordance between the imputed and actual genotypes for each wheat or barley line averaged over the occurrences of that sample within the 100 iterations.

As expected, all metrics were the highest when imputing to the set of SNPs tagged at r² ≥ 0.90 and the lowest for those tagged at r² ≥ 0.50 (Table 1). In wheat, only a small decrease in accuracy was observed for most accessions as the size of the tagged SNP set increased (i.e., r² decreased), with reduced accuracy most evident in the bottom 50 accessions (Figure 4). For these accessions, the difference in accuracy (both correlation squared and concordance) between comparisons including and excluding heterozygous genotype calls was almost 10%, suggesting the possibility of high error rates in the heterozygous exome SNP calls for these accessions. About 768 (88.5%) of the wheat accessions had accuracies ≥ 90% with the strictest correlation squared metric (which included heterozygous calls) for the set of SNPs tagged at r² ≥ 0.50. When comparing only homozygous calls, the number of lines above this threshold rose to 866 (99.8%) (Figure 4).

Reduced accuracy when imputing to higher tagged SNP numbers was more pronounced in barley. A difference of 10.8% (from 96.8 to 86%) was observed between the average correlation squared (which included heterozygous calls) for the set of SNPs tagged at r² ≥ 0.90, compared to those tagged at r² ≥ 0.50 (Table 1). As observed in wheat, the inclusion of heterozygous calls reduced the accuracy, particularly when imputing to the set of SNPs tagged at r² ≥ 0.50, again suggesting possible erroneous heterozygous calls in the sequence genotypes (Figure 4). The reduced accuracies observed in barley compared to wheat are also likely to be partly due to the reduced size of reference haplotypes (155 vs. 868). Accuracies in barley are likely to improve if the reference haplotype set is expanded.

To confirm that the selected tSNPs were useful for detecting marker-trait associations, we performed GWAS using phenotype data for awned status (scored as a presence-absence trait) in 355 wheat accessions and defining row type (scored as two- or six-rowed) in 121 barley accessions and the selected tSNPs before and after imputation. The results were compared with GWAS performed using the same phenotypic data and exome SNP genotypes (Figure 5). Significant and completely overlapping GWAS signals were observed for the three analyses performed in both wheat and barley using the different datasets. The significant SNPs in each analysis were associated with genomic regions previously reported to be associated with the traits (Russell et al., 2016; Huang et al., 2020). While the significance of the associated SNPs differed across the three analyses for each trait, the GWAS results show that the selected tSNPs effectively tag common halotype block diversity in globally diverse germplasm.

The final array design comprised 34,481 tSNPs and two additional categories of context-specific SNPs (content summarized in Table 2; full details are in Supplementary Table 1).

The first context-specific category included 2,609 SNPs from the Infinium wheat 90K SNP array (Wang et al., 2014) that were selected based on allele differentiation to tag tetraploid wheat (A- and B-genome) diversity and to clearly delineate tetraploid wheat from other types of wheat, as well as to distinguish tetraploid species and subgroups from one another. The SNPs comprised the following four classes: (1) differentiating SNPs that represent the top 2% F_st values in the study by Maccaferri et al. (2019) between the four subgroups of tetraploid species: wild emmer (Triticum turgidum ssp. dicoccoides), domesticated emmer (T. turgidum ssp. dicoccocum), and durum (T. turgidum) landraces, and durum cultivars; (2) subgroup-specific private SNPs that showed a MAF ≥ 0.1 in one of the subgroups and were either monomorphic or showed a MAF < 0.05 in the other subgroups; (3) subgroup-specific high MAF SNPs that were present at ≥ 0.3 MAF in any one of the subgroups; and (4) neutral SNPs that did not show any signatures of selection, were polymorphic in all subgroups and showed an overall MAF of ≥ 0.4. The ability of these SNPs to reliably differentiate the tetraploid species subgroups as efficiently as the Infinium wheat 90K array is shown in Supplementary Figure 2. The distribution of these SNPs across the A- and B-genomes of wheat is shown in Supplementary Figure 4.

The second category included 1,206 exome SNPs tagging Ae. tauschii (D-genome) diversity present in backcross synthetic derivatives that originated from crosses involving 100 primary synthetic parents, which were selected for phenotypic and genetic diversity among approximately 400 primary synthetics developed at CIMMYT and imported to Australia in 2001. Each of the 100 primary synthetic parents was derived from a different Ae. tauschii accession. SNPs tagging diversity in Ae. tauschii were selected to provide high genome coverage in the D-genome (Supplementary Figure 4). They were also selected to clearly delineate bread wheat from other types of wheat. The SNP comprised two classes: (1) differentiating SNPs that represent the top 2% F_st values between the global diversity wheat and synthetic derivative collections; and (2) D-genome diversity from Ae. tauschii that showed a MAF ≥ 0.1 in the synthetic derivative collection and MAF ≤ 0.1 in the globally diverse wheat collection. The ability of these SNPs to reliably differentiate synthetic wheat from common wheat as efficiently as the Infinium wheat 90K array is shown in Supplementary Figure 3.

The final category included linked SNPs for key breeding traits and SNPs linking major germplasm resources genotyped with different technologies. In total, 457 wheat and 178 barley SNPs corresponded to published trait-linked markers as well as 109 SNPs associated with agronomically important genes reported in published GWAS studies (Sun et al., 2017; Wang et al., 2017) (Supplementary Table 1). Another 614 SNPs provide a direct link to 19,778 GBS genotyped domesticated barley accessions (Milner et al., 2019).

A limitation of hybridization-based genotyping arrays is that their oligonucleotide probes hybridize both to the targeted locus and its homologs and paralogs, if present (Cavanagh et al., 2013; Wang et al., 2014). Consequently, the ratio of allele-specific fluorescent signals observed for an assay depends on the copy number of the locus in the genome, with increasing copy number reducing the allele-specific fluorescent signal ratio and separation of SNP allele clusters. Further, SNP assay scorability and genotype calling can be confounded by the presence of mutations that modify oligonucleotide annealing such that different cluster patterns are observed across germplasm (Wang et al., 2014). An ideal assay design for a hybridization-based genotyping array is therefore an oligonucleotide probe that binds at only one locus in the genome and has no known nucleotide variation underlying the probe hybridization site. Theoretically, this should ensure three distinct clusters corresponding to the genotypic states (REF, HET, and ALT) expected from a single-copy biallelic SNP. The increasing availability of genomic resources now allows for this historical problem to be addressed. Hence, we used the combination of reference genome assemblies and genotypic data for large globally diverse wheat and barley collections to specifically target the design of single copy biallelic SNP assays.

For the purpose of evaluating the performance of the array, the wheat and barley diversity populations were used to define cluster positions for SNP genotype calling. The vast majority (98%) of the 39,654 SNP assays on the array produced scorable cluster patterns when hybridized with a barley or wheat sample; about 91% (12,949/14,261) of the barley and 83% (20,090/24,598) of the wheat SNP assays could be reliably scored as single-copy biallelic markers, with the REF and ALT clusters having theta values close to 0 and 1 in GenomeStudio SNP plots (Figure 6). While the remaining SNP could typically be reliably scored as biallelic markers, they showed cluster compression indicative of multiple loci. Few assays showed complex clustering patterns, indicating the success of designing probes without any underlying polymorphism. About 5 and 7% of wheat and barley assays showed a clustering pattern typical for the presence of a null allele. The occurrence of assays not behaving as single-copy biallelic markers reflects current knowledge gaps for structural variation in the genomes of wheat and barley including both copy number variation and presence-absence variation (Wang et al., 2014; Balfourier et al., 2019; Walkowiak et al., 2020).

The concordance between called and actual genotypes was exceptionally high for both wheat and barley. The average genotype concordance and correlation squared were 99.5 and 98.1%, respectively, in wheat when heterozygous genotype calls were excluded, and 97.6 and 95.7%, respectively, when heterozygous calls were included. Similarly, 99.8% concordance and 99.2% correlation squared were observed in barley when heterozygous calls were excluded, and 98.2 and 97.2% were observed with heterozygous calls included. The average missing data rates were 4.8 and 3.8% in wheat and barley, respectively.

The design process specifically aimed to select species-specific SNP probes and thus it should be theoretically possible to jointly hybridize a wheat and barley sample to the same bead chip array (dual hybridization) without the loss of genotype calling accuracy. Cross-hybridization between species is expected to confound genotype calling accuracy by creating shifts in SNP cluster positions and/or complex clustering patterns that cannot be easily scored.

To evaluate the assay performance of a dual hybridization, samples from the InterGrain commercial barley and wheat breeding programs were used to define cluster positions and call SNP genotypes for 576 dual hybridization assays. The same samples were also assayed in single sample hybridization assays to enable genotype calling accuracy between dual and single hybridization assays to be directly compared.

Most of the barley and wheat SNPs in dual hybridization assays produced scorable cluster patterns. Shifts in cluster positions were observed, which indicated either that some oligonucleotide probes showed a degree of cross-species hybridization or that deviation from the standard amount of sample DNA (200 ng per sample) recommended for the bead chip assay affected signal-to-noise. Through empirical testing, we found that the quantity of genomic DNA per sample was a major factor causing shifts in cluster position (data not shown) and could be minimized by adjusting the input DNA for each sample to match the ratio of the genome size for each species; e.g., 200 ng barley DNA and 600 ng wheat DNA; the bread wheat genome is about three times larger than that of barley.

For the purpose of assessing genotype calling accuracy for dual hybridization assays, only SNPs that revealed polymorphism among the 576 wheat and barley samples assayed were considered. Of the 9,826 barley and 9,118 wheat SNPs showing polymorphism, the vast majority were easily scored as biallelic markers and had good cluster separation, indicating that oligonucleotide probe cross-species hybridization was minimal (Figure 7). The average concordance between genotypes calls for the same wheat and barley samples in single and dual sample hybridization assays were 99.9, 96.7, and 99.8%, for the REF, HET, and ALT alleles, respectively. The average missing data rate across the wheat and barley samples was similar for both assay types, with 4.7 and 2.0% in dual and single hybridization assays, respectively.