Genomic Evaluations

1
Genomic Evaluations
2
How the system works

• Studs and breeds nominate animals through AIPL
  web site
• Hair, blood, semen, or extracted DNA sent to 1 of
  4 Labs
• Genotypes sent to AIPL monthly
• Monthly evaluation updates released on the first
  Tuesday of most months
• Official evaluations updated only at tri-annual
  traditional runs (except for C P bulls)

3
Pedigree Nomination

• Studs may submit pedigree and nominate in batch
  files
• Pedigree for CAN, AUS, GBR automatically
  collected from web sites
• Nomination expected by the time sample arrives at
  lab
• Sample ID reported at nomination must match ID on
  sample at lab

4
Conflict processing

• Parent-progeny conflicts detected
• Sex and breed checked
• Conflicts reported to lab and requester for
  resolution
• Pedigree changes automatically update genotype
  usability
• Foreign pedigree updates not automatic

5
Changes in April

• Deviations of predictor cows adjusted to be like
  bulls with similar reliability to improve their
  contribution to accuracy
• Genotypes of dams of genotyped animals imputed to
  add predictor animals
• Sum of genomic relationships of each animal with
  the predictor animals used to improve estimation
  of Reliability

6
Imputation

• Determine an animals genotype from genotypes of
  its parents and progeny
• Genotype separated into sire and dam
  contributions. Identifies the allele on each
  member of a chromosome pair

7
O-Style Haplotypeschromosome 15
8
Imputation (cont.)

• Inheritance of haplotypes tracked
• Accuracy of imputation improves with number of
  progeny
• Crossovers during meiosis contribute to
  uncertainty

9
Imputation Plans

• Add separate genomic indicator code (probably 3)
  to cow format 105 to identify imputed cows.
  Already are identified in XML files.
• 3K genotypes will be imputed to 50K, chip type
  code to be added to XML
• No authorization to release evaluations on
  imputed bulls. 15 HO bulls have 5 genotyped
  progeny whose dams are genotyped.

10
Genotyped Holstein by run
Run Date Old Old Young Young Total
Run Date Male Female Male Female Total
0904 7600 2711 9690 1943 21944
0906 7883 3049 11459 2974 25365
0908 8512 3728 12137 3670 28047
0910 8568 3965 13288 4797 30618
1001 8974 4348 14061 6031 33414
1002 9378 5086 15328 7620 37412
1004 9770 7415 16007 8630 41822
1005 9958 7940 16594 9772 44264
Animals with traditional evaluation Animals
with no traditional evaluation
11
Cow Adjustment

• Evaluations of elite cows biased upward
• Cutoff studies showed little benefit from
  including cows as predictors
• Reducing heritability would reduce the problem
  but industry is reluctant to do so
• Adjustment of cow evaluations implemented

12
SD of Cow Deviation from PA
13
Mean of Cow Deviation from PA
14
Cow Adjustment Procedure

• Deregressed Mendelian Sampling (MS)
• (PTA-PA) / f(REL)
• Adj. MS .84MS - 784
• Adj. PTA f(REL)(Adj. MS PAn) (1- (REL)PAn)

f(REL) fraction of PTA from own records and
progeny
15
Effect of Adjustment on Holstein
Bias Bias Bias Regression Regression Regression Gain REL Gain REL Gain REL
No Yes Diff No Yes Diff No Yes Diff
Milk (lb) -75.3 -27.9 47.4 .93 .90 -.03 29.5 32.5 3.0
Fat (lb) -5.7 -2.9 2.8 .98 .97 -.01 34.0 37.1 3.1
Protein (lb) -0.2 0.8 1.0 .90 .97 .07 25.0 27.1 2.1
Fat () 0.0 0.0 0.0 .97 .99 .02 49.8 52.4 2.6
Protein () 0.0 0.0 0.0 .87 .88 .01 38.8 41.5 2.7
16
Effect of Adjustment on Jersey
Bias Bias Bias Regression Regression Regression Gain REL Gain REL Gain REL
No Yes Diff No Yes Diff No Yes Diff
Milk (lb) -44.0 81.5 125.5 .99 .99 .00 10.8 19.6 8.8
Fat (lb) -7.3 7.9 15.2 .78 .84 .06 9.4 18.2 8.8
Protein (lb) 1.7 4.3 2.6 .86 .90 .04 4.1 12.7 8.6
Fat () 0.0 0.0 0.0 .90 .95 .05 29.9 37.6 7.7
Protein () 0.0 0.0 0.0 .87 .93 .06 24.8 34.2 9.4
17
Cow Adjustment Summary

• Increased reliability of genomic predictions
• Genomic evaluations of the top cows, top young
  bulls, and top heifers decreased
• Among bulls, foreign bulls with a high proportion
  of genotyped daughters had largest changes
• Adjusted PTA reported in XML traditional fields

18
Reliability for young HO Bulls
N 15,226
19
Reliabilities for HO born 2005
No Traditional Evaluation No Traditional Evaluation With Traditional Evaluation With Traditional Evaluation
Trait Male Female Male Female
N 15226 7536 752 3191
Milk (lb) 73.9 73.7 85.8 77.9
Protein (lb) 73.9 73.7 85.8 77.8
PL (months) 64.0 63.6 70.1 67.0
SCS 69.7 69.5 78.1 73.0
DPR () 61.6 61.2 66.5 64.6
PTAT 70.4 70.1 78.3 74.5
Sire CE 64.9 61.7 80.8 63.5
Daughter CE 60.2 59.0 69.5 61.8
Sire SB 59.8 58.7 66.2 59.6
Daughter SB 58.3 57.6 64.9 59.6
Net Merit () 68.6 68.3 77.8 72.0
20
Accommodating chip diversity

• Impute to higher density
• Calculate effects for all high density SNP
• Mechanism for accounting for loss in accuracy due
  to imputation error needed
• Percent missing genotypes
• Only observed genotypes stored in database
• Evaluations labeled as to source of genotype

21
Illumina 3K chip

• SNP chosen
• 3072, evenly spaced
• Some Y specific SNP
• 90 SNP for breed determination
• Expect to impute genotypes for 43,382 SNP with
  high accuracy
• Expect breeds to use 3K chip to replace
  microsatellites for parentage verification
• Breeds allowed to genotype bulls for parentage
  only

22
Proposed stud use of 3K chip

• Accuracy adequate for first stage screening
• HD genotyping reserved for bulls acquired.
• Confirm ID
• Second stage selection
• Genotyping of more candidates
• Genotype remaining CDDR predictor bulls to meet
  or exceed EuroGenomics reliabilities

23
HD chip

• Proposed 860K SNP include current 43,382 so can
  replace 50K chip in current evaluations
• 3,000 genotypes at HD may be adequate to support
  imputation of HD from current 50K SNP
• Expected gain in Rel lt 2
• May allow HO genotypes to contribute to accuracy
  of JE BS genomic evaluations

24
HD chip (Cont.)

• Could share cost of HD genotyping with Europe to
  get more animals to improve accuracy of
  imputation
• Trend is toward higher densities
• Continued genotyping at 50K may be shortsighted
• May allow reduction in polygenic effect giving
  increased accuracy

25
Will data recording survive?

• Progeny test no longer required to market bulls
• In 2013, new entrants may have no data collection
  expense
• Loss in accuracy of SNP effect estimates occurs
  over time
• How much data is needed?

26
What replaces the PT program?

• G bulls will have thousands of daughters in their
  early traditional evaluations
• Milk recording is justified for management
  information
• Type data may come from breeder herds because
  they use G bulls
• Data on new traits will have to be paid for

27
Data into National Evaluations

• Progeny test herds could become data supply herds
• Data acquisition could be supported by a fee
  based on animals receiving a genomic evaluation
• Plan must be perceived as fair by all industry
  players
• Quality certification model could apply

28
Questions

• How can accuracy of evaluations from EuroGenomics
  be exceeded?
• Should young bull purchases be based on 3K
  genotypes?
• How will continued flow of data into genetic
  evaluations be assured?

29
Questions from Bob
30
Will there be a code on GPTAs to distinguish
between genotyped and imputed animals?

• Genomic indicator code of 3 planned for imputed
  cows in format 105 in August
• Already designated in XML files

31
From which EU countries are cow proofs used in
genomics? (All, health, type?)

• Cow evaluations for milk, fat, and protein are
  collected from
• NLD
• DEU
• FRA
• GBR
• ITA
• DNK

32
What caused the DPR changes from Dec Feb
April? Can it happen again?

• The traditional DPR PAs for some foreign bulls
  were incorrect in Feb
• May have been due to missing the dam or MGS
• We have increased checking for missing pedigree

33
What is the difference between selection index
(US) and blending of proofs (CAN)?

• Selection Index combines
• Genomic
• Traditional
• Traditional computed using only genotyped animals
• Theoretically justified
• DGV includes all information when both parents
  genotyped
• Used in most countries

34
What is the difference between Selection Index
(US) and Blending of proofs (CAN)? (Cont.)

• Blending combines
• Genomic
• Traditional
• Weighted by reliability
• Simple to explain

35
What are the criteria for an animal to be
included in the reference population for genomics?

• Traditional Rel gt PA Rel

36
What other factors can change SNP effect
estimates, beyond adding new animals to the
reference population?

• New traditional evaluations tri-annual runs
• Insufficient iterations in previous run
• Change in SNP used

37
Why do genomic evaluations change?

• Reference population animals are added
• Changes in traditional PTA cause genomic
  evaluation to change particularly for high
  reliability bulls
• Small changes due to filling in missing SNP
  genotypes when possible

38
How much do the SNP effect estimates change
Trait SNP effect differences from April to May SNP effect differences from April to May Maximum
Trait Mean Std. Dev. Maximum
Milk (lbs) 0.42 0.38 24.5
Fat (lbs) 0.02 0.01 1.1
Protein (lbs) 0.02 0.01 0.8
Fat () 3.9E-5 3.4E-5 9.1E-4
Protein () 1.9E-5 1.6E-5 4.1E-4
39
Are reliability calculations different in the US
vs. EU? And, why are reliability values similar
with a large discrepancy in the number of
predictor bulls? (9,000 vs. 17,000)

• 8,000 cows also contribute to accuracy
• US Rel is adjusted to reflect gains from cutoff
  studies

40
Can sire proofs be imputed? And, is this likely
to happen?

• Genotypes of bulls can be imputed
• Only 15 non-genotyped HO bulls with 5 genotyped
  progeny with genotyped dams
• May be approved for bulls controlled by
  participating studs

41
Will there be an adjustment made to type in
August?

• No

42
Is it possible to estimate the variation in the
offspring for NM when two genotyped animals are
mated?

• Yes, sum absolute value of SNP effects weighed
  as
• Parents both 0 or 2, weight 0
• Parents 0 and 2, weight 1
• 1 or both parents 1, weight 2
• Weights are max difference in progeny genotypes

43
Making genotyped and non-genotyped cows more
comparable

• High priority research area
• Reduce h2
• Add herd x dam interaction
• Differential adjustment by herd

44
Will AIPL continue to impute cows during each
genomic evaluation?

• Yes

45
Explanation of changes in SHOTTLE evaluation from
Jan. to April
January January April April
Trait Traditional Genomic Traditional Genomic
Milk (lb) 1597 1784 1292 1399
Protein (lb) 45 51 38 40
PL (months) 3.0 6.2 3.6 4.4
Net Merit () 529 729 507 551
46
Why were cows in Advantage herds with no
preferential treatment adjusted?

• All genotyped cows were adjusted in the same way
• The maternal component of PA was adjusted
• Investigating if accuracy can be improved by
  adjusting each herd based on its own average
  PTA-PA

47
What changes to the imputation process were made
in May?

• Maternal grandparents were checked for haplotypes
  where parents were not available
• Current allele frequencies replaced base
  population frequencies for unknown genotypes

48
How do you incorporate various chip sets (ex.
3K, 50K, 700K, 850K) into a single genomic
evaluation? And, what level of imputing will take
place?

• Lower densities will be imputed to highest
  density
• If the larger HD chip does not include all the
  SNP of the smaller one, then combined set must be
  imputed or some SNP ignored

49
What will be the gain in accuracy from going from
50K to 850K?

• lt2 increase in reliability

50
What are the biggest changes and challenges after
March 2013 when anyone can get a genomic
evaluation of a bull?

• Maintaining support for data collection for
  genetic evaluations

51
What are the relative weights given to SNP
information for the GPTAs of 1st lactation cows,
1st crop bulls, and 2nd crop bulls?

• Proportional to daughter equivalents (DE)
• DE kRel/(100 Rel)
• Calculate DE at each stage of evaluation
• DEG DEtotal - DEPA

52
How is a DGV calculated?

• ? SNP effects base
• Polygenic effect not included

53
Do SNP estimates change based on family?

• No, SNP effect is change in PTA from having an A
  allele instead of a B allele (substitution effect)

54
How can 99 Reliability bulls change between runs?

• Traditional evaluations change
• High reliability evaluations force SNP effects to
  adjust to equal evaluation
• Possible because more SNP effects than predictor
  animals