Implementation of
genomic evaluations for complex models at a commercial scale
Single-step genomic
evaluation
Our group pioneered the methodology of single-step genomic evaluation. It is a BLUP where the numerator relationship matrix A is replaced by matrix H that combines pedigree and genomic relationships. Single-step results in dramatic simplification of evaluation procedures combined with superior accuracy, ability to use any model and speed similar to that of BLUP. With the APY algorithm that allows for inversion of genomic relationship matrix of any size, there is no longer a limit on the number of genotyped animals. Most of the current projects look at some aspects of the genomic evaluation usually using commercial data sets.
Genomic evaluation of purebreds and crossbreds
Genomic predictions of crossbreds based on purebreds do not seem to be as accurate as predictions for purebred. Use of crossbred data does not seem to improve the accuracy of purebreds much, and putting all breeds together seem to give problematic results, as evidenced by little information coming out from such evaluations. In general, the genomic selection depends on efficient estimation of clusters of chromosome segments. Our hypothesis is that such segments are different in purebreds and crossbreds.
Potential negative effects of genomic selection
Selection on an index of traits causes changes on unselected but correlated traits. With genomic selection we see higher gains for selected traits but probably losses on other traits. The losses may be initially invisible as they are partially compensated by improved management and if they are part of a selection index. Our research is on what gains and what loses. In the end, we need resilient animals that produce well under good management but do not die during temporary challenges.
Changes in genetic
parameter over time
Intensive genomic selection can lead to changes in genetic parameters. In a study in pigs, the additive variance for growth dropped to one half, and the antagonistic correlation between growth and fertility increased. It is not clear whether the changes are artifacts of methodology, due to the Bulmer effect, or is something else acting here. We try to carry more similar studies across species. The studies are hampered by expensive computations.
Dimensionality of genomic information and persistence of
GEBV
The number of chromosome segments seem to be small in farm populations, between 5k and 15k. If the segments are very well estimated, the predictions can be pretty good across generations, unless there is substantial dominance and epistasis. We test whether the persistence increases with more data, for a variety of traits.
Single-step GWAS for large data with p-values
A classical GWAS is applicable to simple models, which may reduce efficiency of QTL detection. We extended GWAS to single-step where any model can be used. Latest refinements allow for computation of p-values. The current work is increasing computational efficiency to allow for unlimited data sets, including with sequence data.
Understanding artifacts and nature of GWAS
Many studies with small data seem to identify lots of QTLs while studies with large data identify just a few. Also, Bayesian methods can identify extremely large signals that in the end often are identified as either spurious or greatly inflated. We try to explain such paradoxes. Large LD due to small population size in farm populations causes QTL signals to be diluted. Manhattan plots contain responses not only due to QTL but also due to tags of chromosome segments of the most popular animals. We try to identify a QTL profile in Manhattan plots that accounts for effective population size fit such a profile in GWAS.
Climate change and heat stress in dairy cattle, beef
and pigs
Whenever there is a heat stress, for example in Southeastern USA, some animals
perform satisfactorily but some perform poorly. We developed a methodology to
study genetics of heat tolerance using easily available weather records. This
project has been described briefly in a UGA
extension letter. Our studies indicate that continued selection for
performance in moderate climates makes cow less heat
tolerant for production and particularly reproduction. Analyzes of US national
data for Holsteins indicated negative trends for heat tolerance. We extended
heat stress studies to beef and pigs. In fast growing animals, the methods can
be used to select differential parents for hot and mild/cold season.
Better modeling of
competitive/associative effects
Total production of animals in a pen, like pigs or chicken, depends not only on individual performance but on dynamics of interactions among the animals. Selection of animals for individual performance may have produced aggressive animals that do not function well in groups, causing reduced growth, injuries and even deaths. Current models based on original work in trees seem to produce ambivalent results. We are trying to find better models based on behavior of specific species.
These and other projects can benefit from your collaboration. If you are looking for a graduate school or a place for a sabbatical, please consider the University of Georgia.
last updated May 25, 2021