Table 1.
Summary of core and supporting metrics to describe marker quality.
Fig 1.
Overview of the marker quality metrics.
Core metrics capture the most critical information relating to marker performance, accuracy and usefulness. Supporting metrics are needed in the calculation of the core metrics and/or capture other important information, but are not routinely required in making breeding decisions.
Fig 2.
Illustration of the evolution of a QTL.
A Starting from an ancestral point, mutations in a particular gene accumulate (numbered black bars, representing mutations in B), resulting in new alleles. At some point, a mutation arises which improves a trait, resulting in a donor allele for a QTL (dark/red), distinguishing it from known recipient alleles (light/blue); typically the status of many alleles is unknown (white). C Each mutation (1–21) is a potential marker, and all are found in the same gene, but some are more informative than the others. Comparing the false positive and false negative rates for each mutation allows the determination of which polymorphism gives the most reliable discrimination between the donor and recipient phenotype classes.
Fig 3.
Determining the optimal polymorphism under several evolutionary scenarios.
Depending on when the causative mutation arose (arrowed bars; outwards pointing for mutations conferring favourable or inwards facing for unfavourable alleles respectively), there may be only one favourable allele (A), a small number of alternative favourable alleles and multiple unfavourable alleles (B), or multiple favourable alleles and a few unfavourable (C). In A and B, the derived allele for the QTL is the favourable allele(s); in C it is the unfavourable. In all cases the polymorphism which gives the most accurate classification of donor and recipient status is one which arose in the same lineage and at a similar time to the causal, derived mutation.
Fig 4.
Alternative markers within a QTL region (markers A—E) each have multiple alleles (numbered). Alternative alleles of each marker are found at differing frequencies within a breeding pool. Those markers with a high frequency of the favourable allele in the breeding pool (B, C, D)–and thus low utility—can only distinguish the donor genotype in a small number of breeding backgrounds. By contrast markers A and E have high or very high utility, as they are polymorphic with respect to nearly all target genomes in the breeding pool.
Table 2.
Ideal target values for key metrics.
Fig 5.
Technical metrics for candidate indel markers.
Markers were evaluated within the qDTY4.1 (A) and qNa1L (B) QTL regions. Significant variation was seen for different markers in both QTL regions, showing some markers clearly performed better than others.
Fig 6.
Comparison of performance of PAGE-based marker systems.
The performance of SSRs was compared to that of trait-specific indels. Indels consistently out-performed SSRs, particularly in marker clarity but also in call rate.
Fig 7.
Use of marker quality metrics to determine optimal markers for a QTL.
Comparison of quality metrics for different markers within a QTL region for salinity tolerance, qNa1L, between 37 and 41Mb on the long arm of chromosome 1. Multiple markers from the Illumina Infinium chip (Anonymous SNPs; favourable allele corrected to minimise FNR), QTL-specific SNPs and QTL-specific indels were assessed for their utility (A), false-positive rate (B) and false-negative rate (C) across IRRI germplasm. Anonymous SNPs typically scored poorly on FPR, FNR and Utility, and none scored well on all metrics. QTL-specific SNP and indel markers typically showed low or perfect scores on the FPR, and several markers scored 0% (no misclassified entries) on both FPR and FNR metrics. In addition QTL-specific markers scored far better for utility, indicating wider applicability in breeding. Arrows to the right indicate ideal target values for a new marker.
Fig 8.
Comparison of mean accuracy metrics for diversity SNPs and QTL-specific SNP and indel markers.
Anonymous SNP markers initially have very low scores on both the FPR and FNR. Correcting the assignation of favourable and unfavourable alleles to minimise the FNR improves that score to equivalent levels with the QTL-specific markers, but the FPR remains poor and no benefit is seen for the utility.
Fig 9.
Variation in utility between various QTL for agronomic traits in indica breeding germplasm.
QTL were selected that have diagnostic markers or markers scoring 0% on both FPR and FNR (and thus could be accurately scored). Wide variation in QTL utilities were seen, from near-fixation (utility ~0%) to absent (utility 100%), but most were rare or absent.