What Does ‘Accuracy’ really mean?
“Accuracy” in genetic terms refers to how precisely we can determine the genetic merit of each animal.
The scale of measuring accuracy is normally reported from 0 to 1 (where 1 is equivalent to 100%).
Measuring Genetic Merit
As the true genetic merit of an animal cannot be measured directly it must be predicted from measurement of performance, information on all available relatives (parents, sibling, progeny) and, more recently, from DNA or genomic information (see Table 1 below for some examples).
Table 1 Accuracy of predicting true genetic merit based on information for different sources
Each source of information (performance, pedigree and genomics), contributes to improving the overall accuracy of breeding value prediction. The relative importance of each source of information depends on the heritability of a trait, and how many other sources are contributing. The main value of genomic information is the contribution it makes to the accuracy of estimating breeding values when animals are very young.
Genomic information is also very valuable for predicting breeding values for traits that are difficult to measure and not normally included in selection decisions. Once the breeding value is reasonably accurate, genomic information becomes less useful.
Overall, the accuracy of selection relates directly to the rate of improvement. If we selected on an index with 0.3 accuracy, we would get 30% of the gain compared to selection based on perfect knowledge of breeding value. The accuracy measures how well we can rank animals for true merit on a scale from 0 to 1.
Selecting for Multiple Traits
When selection is on multiple traits, then traits that have the highest accuracy usually drive most of the genetic change. These are the easy to measure traits, such as growth rate or fleece weight or fibre diameter.
Genomics increases the accuracy of selection especially for traits that were poorly measured before, e.g. meat quality, adult wool traits or reproductive rate and parasite resistance. These traits can now be better selected for, since we can predict genomic breeding values with some accuracy for these traits.
Calculating Accuracy of Breeding Values
Accuracy of predicting breeding values is usually calculated within breeds, or within subsets of breeds such as superfine Merinos. It represents then how well we can rank animals within a flock of sheep of the breed type. This is relevant for breeders that want to select mainly within their flock or breed type.
Ranking animals on genetic merit is much easier across breeds or types of breed. We don’t need to do much recording to predict that a fine wool Merino will have a lower fibre diameter than a strong wool Merino. In other words, the accuracy of selection would be much higher if we selected animals across breeds or breed types.
In the Merino breed there is a wide range of phenotypes and genotypes ranging from flocks producing superfine wool with relatively small carcasses, to flocks characterised by larger bodied sheep producing medium to broad micron wool.
Most commercial breeding decisions are made by selecting animals within Merino sub-groups such as: Superfine; Fine and Medium wool categories. Therefore, Sheep Genetics calculates accuracies for use within each group.
Re-aligning Accuracy to Meet Industry Needs
Initial reports on the accuracy of predicting genomic breeding values for Merinos were released at the time of the 2011 Pilot Projects and were based on whole of breed analysis. These accuracies were generally high (up to 70% for fleece weight and fibre diameter traits) and showed clearly that genomic analysis was able to accurately predict differences between rams across the full spectrum of all types of Merino.
However, it became clear when reviewing the results of the second research Pilot Project in early 2012 that the accuracy of genomic predictions were generally too high for predictions when ranking animals within sub-groups or ‘types’.
Therefore, before commencing the third Genomic Pilot Project, it was decided to use accuracies aligned with ASBV accuracies relevant for selection within Merino sub-groups (Superfine, Fine and Medium wool categories). Those accuracies were lower than for whole breed analyses at around 50% for the wool traits.
How Will the Change Affect My Breeding Program?
In December 2012 all Participants in the Genomics Pilot Project were informed of the change from using total flock Merino estimates to sub-flock values for accuracies of breeding values. At that time there was general agreement from all Participants that this was the best approach.
The change had little, if any, effect on the ranking of animals based on inclusion of genomic information.
Although of lower accuracy than initially reported at the time of the second Genomic Pilot Project in 2011, the impact of the new genomic information in breeding decisions within sub-flock groups is still very positive. Table 2 indicates that the accuracy of predicting breeding values for six-month-old Merino rams can increase by around 40% with the addition of the genomic information (+GS) when compared to just using a combination of pedigree and performance data collected up to six months of age.
Using genomic information to select older rams will result in less improvement. As breeders generally use a mix of younger and older rams we predict that breeders starting to use some six-month-old rams should be able to achieve 18% more genetic gain with genomics, based on current accuracies. For breeders who select rams at 18 months of age, the benefit would be around 12%.
As the cost of genomic testing continues to fall and the accuracies of genomic prediction of breeding values continues to improve, the importance of DNA analysis in breeding programs is certain to increase.
Table 2 Summary of the accuracy of predicting true genetic merit based on use of performance and pedigree data (no GS)
and with addition of genomic information (GS)
This guide is provided to help you understand the complexities of breeding profitable yet functional sheep that are right for your business.