AN EXPLANATION OF GENOMIC EPDS

by: John Genho
Livestock Genetic Services


In order to understand how genomic tests can enhance Expected progeny Differences (EPDs), it makes sense to first step back and look at what EPDs represent. EPDs are a prediction of an animal's genetic merit that comes from a statistical analysis. In some ways, we can view this analysis as a black box with certain inputs and outputs. In this case the inputs are a pedigree and phenotypes (such as weaning weights or birth weights). The pedigree helps us understand how animals are related (or share genes in common with each other). The phenotypes measure performance on a subset of these animals. And the black box puts them together to come up with our genetic prediction or EPDs. This has been the model that we have followed for years in our cattle genetic evaluations.

Several years ago, genomics tests came onto the playing field. When they first arrived, we had hopes (and claims in some cases) that they would eliminate the need for EPDs. Many thought that animals would be able to be selected with a single DNA test without the need for pedigrees and phenotypes. In moving towards this goal, we developed two parallel genetic predictions for animals. We kept the traditional EPDs based on pedigree and phenotypes. But we also developed genetic prediction from DNA tests that were completely independent of the pedigrees and phenotypes that we had collected. These parallel systems in many cases created confusion as animal breeders were given different and often conflicting genetic predictions for the same animals. Many breeders either began to discount the predictions from genomics or to discount the EPDs.

To better understand why the predictions were not always the same, it helps to step back and look at what genomic prediction actually entails. It is easy (and incorrect) to assume that DNA tests tell us the truth about an animal that EPDs have been attempting to find. However, this is not the case. The DNA tests that we have used over the past several years are based on genotypes at some limited number of locations across the genome. In some cases this has been 10-15 different locations, but more recently our predictions are based on 50,000 locations in the genome. What genomic prediction entails is trying to associate differences at these locations with differences in phenotypes. However, the gene that is actually causing the differences in the phenotypes is rarely a location that is a part of our test. Instead we are relying on a chance association between the genotyped location and the nearby gene causing the changes in phenotype. Unfortunately, these chance associations change from population to population, so the DNA tests become population specific. And over time, these chance associations can be degraded (again by chance), so it helps to have a pedigree in place for genomic prediction. So when a group with limited pedigrees and phenotypes (ie a genotyping company) attempts genomic prediction, and a group with a lot of pedigrees and phenotypes but no genotypes (ie a breed association) attempts genetic predictions, it really is no surprise that the predictions are different.

The answer to this issue is to get rid of the parallel set of genetic predictions and move to a single genetic prediction that puts all of the information (pedigree, phenotypes, and genotypes) into the black box to produce genomic enhanced EPDs. The advantage to this is that the extensive pedigree and phenotype databases that breed associations have can be combined with genotypes to produce better EPDs. More information is always better (at least in a geneticist's eyes) so this combining of all information is the best system. There are two negatives to this model though. First, it eliminates the idea that genomic prediction can be done without pedigrees and phenotypes. Under this model, the genomic predictions can be better viewed as strengthening the pedigree and phenotypes rather than eliminating the need for them. And the second issue is that we have to build a better black box (or statistical model) to handle genomics in addition to our pedigrees and phenotypes. There has been extensive research and development over the past several years as we have attempted to come up with this new and improved black box.

The model that IBBA is using for its genomic EPDs is called the single step approach or GBLUP. The idea with the single step approach is to use the genotypes to sort out how animals are related to each other. For instance, two half sibling bulls get a 25 percent relationship in traditional genetic evaluations since they each received 50 percent of their genes from their sire. However, they could have received the same 50 percent from the sire, or received none of the same genes from their sire (both extreme but unlikely cases). Genomics can help us establish what their true relationship actually is by comparing the genotypes. In this case, it may be that the bulls were 20 percent related or it could be that they were 29 percent related. In addition to solving these types of issues, there are animals in the pedigree that have no pedigree relationship but have some genomic relationship. This would arise when animals are related to each other beyond the depth of the pedigree. Once we have the relationships more firmly established based on the DNA tests and the pedigree, we are then able to run EPDs based on the phenotypes we have collected.

The advantage to genomic EPDs is the increase in accuracy, especially for animals that are younger and have lower accuracies in traditional genetic evaluations. For instance, for yearling bulls we would expect their accuracies to increase from the 20 percent to 30 percent range up to the 40 percent to 45 percent accuracy.

While there have been bumps along the path of using genomic tests in selection, the incorporation of this information into the IBBA EPDs is a mark of real progress. Furthermore, it will open doors to future use of this data by IBBA breeders as more samples are collected and models are further refined. The future looks bright for DNA tests at IBBA.







Don't forget to BOOKMARK  
Cattle Today Online!