n. gengler 1,2 , g.r. wiggans *,3 , j.r. wright 3 , and t. druet 1,2
DESCRIPTION
Simultaneous accounting for heterogeneity of (co)variance components in genetic evaluation of type traits. N. Gengler 1,2 , G.R. Wiggans *,3 , J.R. Wright 3 , and T. Druet 1,2 1 Gembloux Agricultural University, Belgium 2 National Fund for Scientific Research, Brussels, Belgium - PowerPoint PPT PresentationTRANSCRIPT
2001ASAS/ADSA 2001 Conference (1)
Simultaneous accounting Simultaneous accounting for heterogeneity of for heterogeneity of (co)variance components in (co)variance components in genetic evaluation of type genetic evaluation of type traitstraits
N. GenglerN. Gengler1,21,2, G.R. Wiggans, G.R. Wiggans*,3*,3, , J.R. WrightJ.R. Wright33, and T. Druet, and T. Druet1,21,2
11 Gembloux Agricultural University, Belgium Gembloux Agricultural University, Belgium22 National Fund for Scientific Research, Brussels, National Fund for Scientific Research, Brussels, BelgiumBelgium33 Agricultural Research Service, USDA, Beltsville, Agricultural Research Service, USDA, Beltsville, MD MD
2001ASAS/ADSA 2001 Conference (2)
USDA Type evaluation
Breeds: Ayrshire Brown Swiss Guernsey Jersey Milking Shorthorn Red and White
15 Linear traits and Final ScoreMulti-trait Model
Canonical Transformation Estimation of missing values
2001ASAS/ADSA 2001 Conference (3)
Heterogeneity of (co)varianceVariances and (co)variances assumed
constant across herds and timeHolstein Association accounts for
heterogeneity in final scoreVariance tends to decrease with
increasing herd average final scoreChanges in appraisal program over
time can be a cause
2001ASAS/ADSA 2001 Conference (4)
Methods of HV estimation
Holstein Association estimates phenotypic variance using combination observed variance predicted variance from model including
– mean final score– registry status– number of appraisals for herd-classification
date
Meuwissen proposed simultaneous estimation of variances and breeding values expected to improve accuracy of both
estimates
2001ASAS/ADSA 2001 Conference (5)
USDA HV adjustment system
Apply Canonical transformation to linear traits
Prepare data for 2 models
score = herd-sire + herd-year-season-parity + parity-time_period-age + parity-time_period-stage
var = mean + parity-group_size + parity-herd_mean_final_score + parity-year-season + herd-year-month
2001ASAS/ADSA 2001 Conference (6)
Variance model
Herd-year-month effect in variance model random
Regression of variances within parity toward population mean (fixed effects)
Method R estimation of variance ratios within each EBV iteration
2001ASAS/ADSA 2001 Conference (7)
Computational requirementsComputational requirements
February 2001 Jersey Evaluation Run
number of rounds
time required
(hr)no HV 42 13
with HV 43 26.5
2001ASAS/ADSA 2001 Conference (8)
Iteration
Time increased due to iteration for variance model estimation of variance ratios
Convergence sensitive to quality of the starting values size of unknown parent groups
Ways of reducing processing time parallel processing using a faster computer limiting iteration for variances
2001ASAS/ADSA 2001 Conference (9)
Correlations between HV-adjusted official evaluations
February 2001 official evaluations2497 Jersey AI bulls born 1980+All correlations .989 or higher
2001ASAS/ADSA 2001 Conference (10)
Change in slope of PTA for Change in slope of PTA for Jersey AI bullsJersey AI bulls
Trait
slope
official
slope
HV adj diff
pct
diff
Dairy form 0.27 0.24 -0.036
-15.4
Rear udder height
0.20 0.17 -0.028
-16.2
Front teat place 0.09 0.07 -0.018
-26.4
Strength 0.08 0.07 -0.007
-10.2
Fore udder attach
0.06 0.05 -0.003
-5.7
Teat length 0.02 0.02 0.004
22.5
Udder Depth -0.05 -0.04 0.010 24.8
2001ASAS/ADSA 2001 Conference (11)
Differences between HV-adjusted and official evaluationStandard deviations & mean absolute
values increased as reliabilities increased to 80% decreased slightly for reliabilities of > 90%
Mean differences largest for bulls born from 1985 through 1994 with lowest mean daughters final scores
2001ASAS/ADSA 2001 Conference (12)
Mendelian sampling
Mendelian sampling evaluation minus mean of parent
evaluations
Jersey cows born from 1984 through 1998 regression of SD on birth year
2001ASAS/ADSA 2001 Conference (13)
Mendelian samplingMendelian sampling
Trait
slope
official
slope
HV adj diff
pct
diff
Foot angle -0.0050 -0.0026 0.0024 51
Rump angle -0.0160 -0.0122 0.0038 76
Rear udder height
-0.0050 -0.0011 0.0039 21
Teat placement -0.0097 -0.0055 0.0042 57
Stature -0.0199 -0.0135 0.0064 68
2001ASAS/ADSA 2001 Conference (14)
Interbull trend validation
Trend tests conducted for Stature Udder support
Comparison first parity v. all parities without recent years v. all data
Trend differences within tolerance
2001ASAS/ADSA 2001 Conference (15)
HV model implementation plans
Jersey - May 2001Other breeds by Feb 2002
2001ASAS/ADSA 2001 Conference (16)
Conclusions
Simultaneous estimation of BV & Variances possible
Requires substantially more computer time
Improves stability of Mendelian Sampling Variance over time
HV adjusted EBV enable more accurate selection decisions