Files
Abstract
Catfish farming is the largest segment in the US aquaculture business and among other
topics, the implementation of genomic selection has been recently investigated. Using genomic
information improved predictive ability by 28% for harvest weight and up to 36% for carcass traits
compared to traditional evaluation. This demonstrates the benefit of genomic selection for the US
catfish breeding program. Such improvements have made the use of genomic information widely
adopted across many livestock and aquaculture species. With this rapid adoption, the number of
genotyped animals has been steadily increasing, especially in the US dairy and beef industries.
With a large number of genotyped animals, genomic evaluations may be challenging and indirect
predictions (IP) can be a useful tool providing fast interim evaluations for young genotyped
animals. Further, IP can be used as genomic prediction for unregistered animals not included in
official evaluations. When genomic best linear unbiased prediction (GBLUP) or single-step
GBLUP (ssGBLUP) are the methods of choice for genomic evaluations, IP can be obtained based
on single nucleotide polymorphism (SNP) effects that are backsolved using genomically estimated
breeding values (GEBV). With large number of genotyped animals, IP can be reliably obtained
from (ss)GBLUP either by using direct inversion of G or by using the algorithm for proven and
young (APY) as long as GEBV are from a previous (ss)GBLUP evaluation. Further, in purebred
beef cattle populations, a sample of at least 15,000 animals representing the whole genotyped
population may also provide reliable IP. To make use of IP, it is important that a measure of
accuracy that is comparable to the GEBV accuracy is available. Under (ss)GBLUP, IP accuracy
can be obtained by backsolving prediction error covariance (PEC) of GEBV into PEC of SNP
effects. The computational cost of PEC computations is prohibitive with large number of animals
and using a subset of animals to approximate it is desirable for large scale evaluations. It is possible
to reduce the number of genotyped animals in PEC computations, but accuracies may be
underestimated and fine tuning is still required to scale accuracies of indirect predictions up to
accuracies of GEBV
topics, the implementation of genomic selection has been recently investigated. Using genomic
information improved predictive ability by 28% for harvest weight and up to 36% for carcass traits
compared to traditional evaluation. This demonstrates the benefit of genomic selection for the US
catfish breeding program. Such improvements have made the use of genomic information widely
adopted across many livestock and aquaculture species. With this rapid adoption, the number of
genotyped animals has been steadily increasing, especially in the US dairy and beef industries.
With a large number of genotyped animals, genomic evaluations may be challenging and indirect
predictions (IP) can be a useful tool providing fast interim evaluations for young genotyped
animals. Further, IP can be used as genomic prediction for unregistered animals not included in
official evaluations. When genomic best linear unbiased prediction (GBLUP) or single-step
GBLUP (ssGBLUP) are the methods of choice for genomic evaluations, IP can be obtained based
on single nucleotide polymorphism (SNP) effects that are backsolved using genomically estimated
breeding values (GEBV). With large number of genotyped animals, IP can be reliably obtained
from (ss)GBLUP either by using direct inversion of G or by using the algorithm for proven and
young (APY) as long as GEBV are from a previous (ss)GBLUP evaluation. Further, in purebred
beef cattle populations, a sample of at least 15,000 animals representing the whole genotyped
population may also provide reliable IP. To make use of IP, it is important that a measure of
accuracy that is comparable to the GEBV accuracy is available. Under (ss)GBLUP, IP accuracy
can be obtained by backsolving prediction error covariance (PEC) of GEBV into PEC of SNP
effects. The computational cost of PEC computations is prohibitive with large number of animals
and using a subset of animals to approximate it is desirable for large scale evaluations. It is possible
to reduce the number of genotyped animals in PEC computations, but accuracies may be
underestimated and fine tuning is still required to scale accuracies of indirect predictions up to
accuracies of GEBV