Science (Communication) and Wikipedia - Potentials and Pitfalls
Genetic Evaluation of Calving Traits in US Holsteins
1. 200
6
J.B. ColeJ.B. Cole
Animal Improvement Programs Laboratory
Agricultural Research Service, USDA, Beltsville, MD
jcole@aipl.arsusda.gov
Genetic Evaluation of CalvingGenetic Evaluation of Calving
Traits in US HolsteinsTraits in US Holsteins
2. CSU 2006 – Breeding and Genetics Seminar (2) Cole
200
6
IntroductionIntroduction
A national evaluation was implemented
for calving ease (CE) in August 2002 and
for stillbirth (SB) for Holstein in August
2006.
A calving ability index (CA$) which
includes SB and calving ease (CE) was
developed.
Some challenges with the CE and SB
evaluations remain
3. CSU 2006 – Breeding and Genetics Seminar (5) Cole
200
6
Calving Ease DefinitionCalving Ease Definition
Reported on a five-point scale:
1 = No problem
2 = Slight problem
3 = Needed assistance
4 = Considerable force
5 = Extreme difficulty
Scores of 4 and 5 are combined
4. CSU 2006 – Breeding and Genetics Seminar (6) Cole
200
6
Stillbirth DefinitionStillbirth Definition
Reported on a three-point scale:
Scores of 2 and 3 are combined
1 = calf born alive,
2 = calf born dead,
3 = calf died within 48 h of
parturition.
5. CSU 2006 – Breeding and Genetics Seminar (7) Cole
200
6
Distribution of Stillbirth and CalvingDistribution of Stillbirth and Calving
Ease ScoresEase Scores
7,484,30929,320348,6775,348,0291,758,283Total
96,0871,27232,19638,92923,6905
207,2421,74037,851108,03759,6144
633,0293,35370,522375,203183,9513
738,8532,53749,858482,720203,7382
5,809,09820,418158,2504,343,1401,287,2901
Total3210
CalvingEaseScore
Stillbirth Score
6. CSU 2006 – Breeding and Genetics Seminar (8) Cole
200
6
Stillbirth Records by LactationStillbirth Records by Lactation
0
100
200
300
400
500
1980 1990 2000
Birth Year
NumberofRecords(thousands)
3
2
1
7. CSU 2006 – Breeding and Genetics Seminar (9) Cole
200
6
Detecting Stillbirth Data ErrorsDetecting Stillbirth Data Errors
0
5
10
15
1980
1983
1986
1989
1992
1995
1998
2001
2004
Birth Year
Percentage
%SB
%DB
8. CSU 2006 – Breeding and Genetics Seminar (10) Cole
200
6
Data and EditsData and Edits
7 million SB records were available for
Holstein cows calving since 1980
Herds needed ≥10 calving records
with SB scores of 2 or 3 for inclusion
Herd-years were required to include
≥20 records
Only single births were used (no
twins)
9. CSU 2006 – Breeding and Genetics Seminar (11) Cole
200
6
Sire-MGS Threshold ModelSire-MGS Threshold Model
Implemented for calving ease (Aug
2002) and stillbirth (Aug 2006)
Sire effects allow for corrective
matings in heifers to avoid large
calves
MGS effects control against
selection for small animals which
would have difficulty calving
10. CSU 2006 – Breeding and Genetics Seminar (12) Cole
200
6
Genetic Evaluation ModelGenetic Evaluation Model
A sire-maternal grandsire (MGS)
threshold model was used:
• Fixed: year-season, parity-sex, sire and
MGS birth year
• Random: herd-year, sire, MGS
(Co)variance components were
estimated by Gibbs sampling
• Heritabilities are 3.0% (direct) and 6.5%
(MGS)
ijklnoprnplonlkjiijklnopr emsBMBSPSYShyy +++++++= ijklnoprnplonlkjiijklnopr emsBMBSPSYShyy +++++++= ijklnoprnplonlkjiijklnopr emsBMBSPSYShyy +++++++=
11. CSU 2006 – Breeding and Genetics Seminar (13) Cole
200
6
Trait DefinitionTrait Definition
PTA are expressed as the expected
percentage of stillbirths
Direct SB measures the effect of the calf
itself
Maternal SB measures the effect of a
particular cow (daughter)
A base of 8% was used for both traits:
Direct: bulls born 1996–2000
Maternal: bulls born 1991–1995
12. CSU 2006 – Breeding and Genetics Seminar (14) Cole
200
6
Phenotypic Trend for StillbirthsPhenotypic Trend for Stillbirths
0
2
4
6
8
10
12
14
1980
1983
1986
1989
1992
1995
1998
2001
2004
Birth Year
%Stillbirth
Heifers
Cows
All animals
13. CSU 2006 – Breeding and Genetics Seminar (15) Cole
200
6
Genetic Trend for StillbirthsGenetic Trend for Stillbirths
5.0
6.0
7.0
8.0
9.0
10.0
1980 1985 1990 1995 2000
Birth Year
%SBH
Direct
Maternal
14. CSU 2006 – Breeding and Genetics Seminar (16) Cole
200
6
Distribution of PTADistribution of PTA
0
5
10
15
20
25
30
35
40
6.1-6.56.6-7.07.1-7.57.6-8.08.1-8.58.6-9.09.1-9.59.6-10.010.1-10.510.6-11.011.1-11.511.6-12.0
>12.0
%SBH
PercentofScores
Direct
Maternal
15. CSU 2006 – Breeding and Genetics Seminar (17) Cole
200
6
Distribution of ReliabilitiesDistribution of Reliabilities
0
5
10
15
20
25
30
35
40
45
50 41-45
46-50
51-55
56-60
61-65
66-70
71-75
76-80
81-85
86-90
91-95
96-99
Reliability
Percentage
Direct
Maternal
16. CSU 2006 – Breeding and Genetics Seminar (18) Cole
200
6
Dystocia and StillbirthDystocia and Stillbirth
Meyer et al. (2001) make a strong argument for
the inclusion of dystocia in models for SB
Difficulty of interpretation - formidable
educational challenge
Interbull trait harmonization - none of the
March 2006 test run participants included
dystocia in their models
Changes in sire and MGS solutions on the
underlying scale between models were small
17. CSU 2006 – Breeding and Genetics Seminar (19) Cole
200
6
Evaluation ConclusionsEvaluation Conclusions
Reliabilities for SB averaged 45%
versus 60% for CE
Phenotypic and genetic trends
from 1980 to 2005 were both small
An industry-wide effort is
underway to improve recording of
calf livability
18. CSU 2006 – Breeding and Genetics Seminar (20) Cole
200
6
Index DataIndex Data
7 million SB records were available for
Holstein cows calving since 1980
Calvings with unknown MGS were
eliminated for VCE
Records with sire and MGS among the
2,600 most-frequently appearing bulls
were selected
19. CSU 2006 – Breeding and Genetics Seminar (21) Cole
200
6
Data (cont’d)Data (cont’d)
Herds needed ≥10 calving records with SB
scores of 2 or 3 in the database to be
included
Herd-years were required to include ≥20
records and only single births were used
Inclusion of all records for a cow was not
guaranteed
The final dataset included 2,083,979 calving
records from 5,765 herds and 33,304 herd-
years
20. CSU 2006 – Breeding and Genetics Seminar (22) Cole
200
6
SamplingSampling
Six datasets ofSix datasets of ~250,000~250,000 records each wererecords each were
created by randomly sampling herd codescreated by randomly sampling herd codes
without replacementwithout replacement
Datasets ranged fromDatasets ranged from 239,192239,192 toto 286,794286,794
observations, and all averagedobservations, and all averaged 7%7% stillbirthsstillbirths
A common pedigree file was used to facilitateA common pedigree file was used to facilitate
comparisons between sire and MGS solutionscomparisons between sire and MGS solutions
23. CSU 2006 – Breeding and Genetics Seminar (25) Cole
200
6
Genetic Correlations Among SB and CEGenetic Correlations Among SB and CE
Trait
CE SB
Direct Maternal Direct Maternal
CE
Direct 1.00 0.46 0.67 0.25
Maternal 1.00 0.29 0.63
SB
Direct 1.00 0.28
Maternal 1.00
24. CSU 2006 – Breeding and Genetics Seminar (26) Cole
200
6
Economic AssumptionsEconomic Assumptions
Newborn calf value
Expenses per difficult birth (CE ≥4)
$450 for females
$150 for males
$75 labor and veterinary
$100 reduced milk yield
$75 reduced fertility and longevity
1.5% chance of cow death ($1800)
25. CSU 2006 – Breeding and Genetics Seminar (27) Cole
200
6
Calving Ability IndexCalving Ability Index
CA$ has a genetic correlation of 0.85 with
the combined direct and maternal CE
values in 2003 NM$ and 0.77 with
maternal CE in TPI
Calving traits receive 6% of the total
emphasis in NM$ (August 2006 revision)
(DCE ) (MCE ) (DSB ) (MSCA$ B )= − − − − − − − −4 8 3 8 4 8 8 8
26. CSU 2006 – Breeding and Genetics Seminar (28) Cole
200
6
Breeds Other Than HolsteinBreeds Other Than Holstein
Brown Swiss economic values are
−6 for SCE and −8 for DCE
• Separate SB evaluations are not
available
• CE values include the correlated
response in SB
Other breeds will be assigned CA$
of 0
27. CSU 2006 – Breeding and Genetics Seminar (29) Cole
200
6
Calving Ease Genetic CorrelationsCalving Ease Genetic Correlations
Service sire above diagonal, daughter belowService sire above diagonal, daughter below
Ctry CAN DNK FRA ITA NLD SWE USA
CAN .87 .81 .70 .80 .86 .75
DNK .84 .93 .77 .86 .96 .90
FRA .80 .80 .74 .84 .91 .88
ITA .58 .59 .85 .60 .70 .61
NLD .89 .81 .79 .59 .89 .79
SWE .75 .82 .89 .78 .69 .86
USA .71 .78 .93 .76 .77 .87
28. CSU 2006 – Breeding and Genetics Seminar (30) Cole
200
6
Stillbirth Genetic CorrelationsStillbirth Genetic Correlations
Service sire above diagonal, daughter belowService sire above diagonal, daughter below
Ctry DNK FIN ISR NLD SWE USA
DNK .85 .82 .67 .92 .70
FIN .82 .77 .64 .82 .65
ISR .67 .70 .73 .80 .66
NLD .82 .77 .60 .65 .63
SWE .88 .92 .65 .73 .64
USA .81 .87 .60 .71 .87
29. CSU 2006 – Breeding and Genetics Seminar (31) Cole
200
6
Brown Swiss Calving EaseBrown Swiss Calving Ease
Service sire correlations above diagonal, daughter belowService sire correlations above diagonal, daughter below
Ctry CHE DEU NLD USA
CHE .83 .81 .68
DEU .61 .77 .67
NLD .89 .76 .79
USA .70 .61 .76
30. CSU 2006 – Breeding and Genetics Seminar (32) Cole
200
6
Index ConclusionsIndex Conclusions
A routine evaluation for stillbirth in US
Holsteins was implemented in August 2006
Direct and maternal stillbirth were included in
NM$ for Holsteins starting in August 2006
August 2006 data were included in the
September 2006 Interbull test run
The US will participate in routine Interbull
evaluations beginning in November 2006
31. CSU 2006 – Breeding and Genetics Seminar (33) Cole
200
6
Recent Calving Ease ResearchRecent Calving Ease Research
32. CSU 2006 – Breeding and Genetics Seminar (34) Cole
200
6
Abnormal Herd-YearsAbnormal Herd-Years
Many herd-years have abnormal
distributions of scores
Two recent approaches to problem
• Eliminate HY based on GoF tests
• Collapse categories when mode > 1
Both strategies improve prediction
of later evaluations by earlier
33. CSU 2006 – Breeding and Genetics Seminar (35) Cole
200
6
An IllustrationAn Illustration
Herds with unusual distributions of data affect evaluations of
bulls
Worst case is when large share of records for a bull are in
one “bad” herd
Herd reporting changes over time
0
20
40
60
80
100
1 2 3 4 5
Calving Ease Scores - Herd 1
ScorebyHerd(%)
Parity 1
Parity 2
0
20
40
60
80
100
1 2 3 4 5
Calving Ease Scores - Herd 2
ScorebyHerd(%)
34. CSU 2006 – Breeding and Genetics Seminar (36) Cole
200
6
Test Edits -Test Edits - χχ22
GoF statisticsGoF statistics
Based on multinomial distributions
Independent of herd size
=
= ∑
2
i i,1 i,2 i,3 i,4 i,5
i 1 i
Log(Multi(N ,n ,n ,n ,n ,n ,P))
GoF
N
35. CSU 2006 – Breeding and Genetics Seminar (37) Cole
200
6
Percentage of Score by Parity In AllPercentage of Score by Parity In All
(AN) and GoF Excluded (AG) Herds(AN) and GoF Excluded (AG) Herds
0
10
20
30
40
50
60
70
80
90
100
1 2 3 4 5
Calving Ease Score
CountsbyHerd-Parity(%)
Parity 1 - AN
Parity 2 - AN
Parity 1 - AG
Parity 2 - AG
36. CSU 2006 – Breeding and Genetics Seminar (38) Cole
200
6
Collapse CategoriesCollapse Categories
The mode for CE scores in a herd is
expected to be 1, but was higher for
nearly 10% of data
Data from herd-years with a mode of 4
or 5 (1.2%) were deleted
A mode of 3 is assumed to indicate that
the scorer normalized the data (middle
score of 3 for an 'average' birth)
37. CSU 2006 – Breeding and Genetics Seminar (39) Cole
200
6
Collapse CategoriesCollapse Categories
Herds with a mode of 2 or 3: scores up to the
mode were changed to 1, and scores greater
than the mode were decreased accordingly
Herd-years with a mode of 3: scores 1-3 all
become 1, scores of 4 are changed to 2, and
scores of 5 are changed to 3
Combining categories lowered the portion of
difficult calvings and increased the impact of
the subsequent goodness-of-fit test
Overall, 6.4% of data were excluded
38. CSU 2006 – Breeding and Genetics Seminar (40) Cole
200
6
ConclusionsConclusions
Exclusion of herds with poor
distributions improves prediction of
future evaluations across birth years
• Correlations across all data increased from .
66 to .68
Herds with poor score distributions
were excluded uniformly across herd
size
Exclusion of herds results in loss of
evaluations for some bulls
39. CSU 2006 – Breeding and Genetics Seminar (41) Cole
200
6
Separate Parity EffectsSeparate Parity Effects
First and later parities currently
modelled as a single trait
cblup90iod only accepts one
threshold trait
Options for bivariate analysis
• Gibbs sampling (thrgibbs1)
• Linearization (airemlf90)
• RR on parity (cblup90iod)
40. CSU 2006 – Breeding and Genetics Seminar (42) Cole
200
6
ResultsResults
RR on a 0-1 parity effect does not
account for heterogeneous variances
GS and AIREML solutions were similar
• GS required more processing time than is
desirable for routine national evaluations
• The impact of the approximation necessary
to linearize the scores is not known
Implementation of a bivariate analysis is
desirable, but challenging
41. CSU 2006 – Breeding and Genetics Seminar (43) Cole
200
6
AcknowledgmentsAcknowledgments
Jeff Berger, Iowa State University
John Clay, Dairy Records Management
Systems
Ignacy Misztal and Shogo Tsuruta,
University of Georgia
National Association of Animal Breeders