Changes to evaluation system (April
2010)
Cow adjustments in genomic
evaluation
By George Wiggans, Tabatha Cooper, and Paul VanRaden
Traditional predicted transmitting abilities (PTAs) are inputs
for genomic evaluation, but cows with high traditional PTAs for yield traits
may be overevaluated. Adjustments were developed to rescale cow PTAs for yield
traits to be more comparable to bull PTAs. Resulting genomic evaluations were
more accurate when the rescaled cow PTAs were included in estimation of marker
effects. Genomic evaluations for the top cows, top young bulls, and top heifers
decreased by about 250 pounds for milk, 8 pounds for fat, and 5 pounds for
protein, whereas genomic evaluations for the top bulls with daughters decreased
only by about 70 pounds for milk, 3 pounds for fat, and 2 pounds for protein.
Adjustments were largest for foreign bulls with a high proportion of genotyped
daughters. The population average of all bulls with daughters decreased
slightly by 40 pounds for milk, 1.5 pounds for fat, and 1.5 pounds for protein;
standard deviations also decreased slightly by about 1%. Correlations between
genomic evaluations before and after the adjustment were 0.997 for bulls with
daughters and 0.990 for cows and young animals. Net merit decreased by as much
as $40 for many of the top cows and young animals because of the yield trait
adjustments. Only yield and component percentage traits were adjusted because
non-yield traits had less variance in females than males and did not benefit
from applying these same methods for PTA adjustment. [See
Wiggans et
al. (2010) for further documentation of methods.]
March 30 update 1: The PTAs computed from
February genotypes and January yield data with the cow adjustments applied are
available for preview to help in understanding differences directly caused by
the change. Files containing Holstein and Jersey evaluations for both young and old bulls are available for download at
ftp://aipl.arsusda.gov/pub/outgoing/UNOFFICIAL_1002_NEW_MODEL.zip.
Evaluations in the preview files were not calculated using the new edit for productive life (PL) described in March 30 update 2.
March 30 update 2: A new edit was
introduced for multitrait PL of cows. Previously,
correlated traits measured during first lactation received credit when
calculating genomic PTAs of genotyped cows. For a few prominent sires, this
created an upward bias in PTA PL because many live, genotyped daughters
received credit in marker effect estimation before information from any of the
culled daughters entered the sire's traditional PL evaluation. The new edit
includes only multitrait PL of genotyped daughters after they reach 36 months
of age so that records from living and culled daughters enter equations at the
same time.
The Document 'Imputation of Cow Genotypes
and Adjustment of PTAs' is available for further explaination
Haplotypes used for imputing
genotypes
By Paul VanRaden, George Wiggans, and Jeff O'Connell
New methods were used for imputing missing genotypes in genomic
evaluation. About 1% of marker genotypes are missing in initial data, and
previously about half of those were imputed using gene content of relatives and
an algorithm that did not use neighboring markers or map location on the
chromosome. Program findhap.f90 was developed to determine the paternal and
maternal haplotypes that sum to the animal's genotype. Simulation results
indicated that the initial rate of 99.0% for correct genotypes was increased to
99.95% after imputing the missing values with haplotypes. The same methods will
allow imputing the much higher proportions of missing data that will result
when lower or higher density marker sets are mixed with current data containing
50,000 markers. More than 1,000 nongenotyped cows had at least 90% of their
single nucleotide polymorphisms (SNPs) correctly imputed using genotypes
from multiple progeny (usually 5 or more). Those imputed genotypes are included
in SNP estimation, and such cows will be listed in evaluation files with the
genomic indicator set to 1 and will have a sample identification of "imputed"
in the XML file. [See
VanRaden
(2010) and
VanRaden
et al. (2010) for further documentation of haplotyping methods.]
Breed-specific
markers
By Tabatha Cooper, George Wiggans, and Katie Olson
Genotypes are checked using breed-specific markers to verify that the
reported breed matches the animal's actual breed as part of routine USDA data
edits. Breed is validated using SNPs that are nearly monomorphic in 1 breed and
have fewer than 30% of animals homozygous for that allele in another breed. A
total of 672 SNPs were selected with approximately equal numbers of SNPs that
were monomorphic for each of 3 breeds (Holstein, Jersey, and Brown Swiss). The
number of SNPs for which the genotype differed from the monomorphic genotype
was counted separately for each breed, and the lowest breed count identified
the breed of the sample. In addition to determining an animal's breed, the
number of breed-specific SNPs can indicate partial contribution of another
breed to the animal. More accurate estimates of breed composition can be
produced using regression coefficients for all 43,385 markers, but that test is
not yet used routinely. The simple SNP test may underestimate the actual
percentage of other breed genes present in the animal. For example, full
siblings with a great-grandparent of another breed may have a range of 12 to 26
SNP conflicts with breed-specific SNPs (5.4 to 11.7%). The tool was developed
primarily to detect breed reporting errors, and breed purity should not rely
solely on the approximately 200 SNPs used in detection. Because of the
variation in inheritance of the 200 breed-specific SNPs, some animals with an
ancestor of another breed may not be detected, and other animals with many
generations of pure breeding may have some SNPs specific to another breed.
Animals with 10 conflicts or more at the breed-specific SNP locations are
reported to the requester (breed association or artificial-insemination
organization) of the genotype. If more investigation is needed, the requester
may ask for additional data from USDA. [For more information on selection of
breed-specific SNPs, see
Wiggans
et al. (2010).]
New fertility records excluded
from sire conception rate
By Duane Norman and Paul VanRaden
The numbers of new records received for evaluations for sire, cow, and
heifer conception rates are considerably lower than in past evaluations. As of
January 2010, fertility data in format-5 (reproductive) records were provided
for most herds by 3 dairy records processing centers (AgriTech
Analytic, AgSource, and Dairy Records Management Systems) and for 6 herds by a
fourth processing center (DHI-Provo). For April 2010 evaluations, records from AgriTech
Analytics continued to be provided for most herds, but records from Dairy
Records Management Systems and AgSource were provided only for cows inseminated
to Genex/CRI bulls. The reduction in additional numbers of format-5 records had
a small limited effect on daughter pregnancy rate evaluations because current
days open are reported in format-4 (lactation) records. However, April genetic
evaluations for cow and heifer conception rates have less increase in
reliability than in January for most bulls. When genetic evaluations for
conception rate are released in the future (possibly August 2010) for cows, any
cow without a breeding record will not receive an evaluation.
Evaluations of sire conception rate from January will remain official
and will not be updated in April until the partial data can be modeled properly
or the unusual results obtained from it can be explained. The input records
caused distortions for individual bulls in the proportion of mates observed to
be not pregnant by later inseminations versus confirmed pregnant by examination.
This caused changes in rank much larger than expected from the number of new
records received. Thus, the format-38 (bull evaluation) file in April will contain sire conception rates from
January. Further work is needed to resolve the data submission issues and the
bias created by partial record submission.
Genotypes from
Switzerland
By George Wiggans
An additional 400 Brown Swiss genotypes were traded with Switzerland
prior to the February genomic update. This increased the size of the reference
population by about 50% and increased the genomic reliabilities.
Improved reliability for genomic
evaluations
By George Wiggans and Paul VanRaden
The contribution to reliability from genomics was modified to consider
the sum of genomic relationships of the animal with each animal in the
predictor population weighted by the reliability of that predictor animal's
evaluation. Previously, the contribution to accuracy from genomics was assumed
to be the same for each animal within a breed. The standard deviation of the
change in reliability was about 2 for young bulls and heifers and 1 for bulls
that had daughters included in their traditional evaluations. [For more
information on the genomic reliability approximation, see
Wiggans
and VanRaden (2010).]