Transcript How does additional information impact accuracy?

How Does
Additional Information
Impact Accuracy?
Dan W. Moser
Department of Animal Sciences and Industry
Kansas State University, Manhattan
http://www.asi.ksu.edu
Additional Information
• Start with a pedigree, possibly a phenotype
• Add:
– Phenotype for a correlated indicator trait
– Progeny test data
– DNA test results
• How much is accuracy of selection
increased by adding each piece of
additional information?
What Drives Genetic Change?
 BV / t 
rBV , EBV i BV
L
Bourdon, 2000
What Drives Genetic Change?
•
•
•
•
Accuracy of Selection
Generation Interval
Genetic Variation
Intensity of Selection
Rate of Genetic Change
• Additional information
– Should increase accuracy
– May increase generation interval
– May reduce intensity of selection
• If only certain animals have additional info
It Often Comes Down To:
• Is the value of the additional information
worth the time and cost required to get it?
• Is the additional information available on
enough animals to maintain selection
intensity?
Correlated Traits
• Birth Weight as a predictor of Calving Ease
– Because Calving Ease is a binary trait
(yes/no), having a correlated continuous
indicator trait can add considerable accuracy
to selection
Correlated Traits
Calf
CE
1
U
2
U
3
U
4
U
5
U
6
U
7
A
8
A
BW
65
70
75
80
85
90
75
95
CE: U = Unassisted, A = Assisted
Unmeasured ERT
• ERT = Economically relevant trait
• When ERT is unmeasured, the genetic
relationship between the ERT and the
indicator trait must be strong
• Otherwise accuracy of selection for the
ERT using only the indicator is low
Accuracy of Correlated Selection
Acc (ERT selection) = Acc (Indicator EPD)
x rG btw ERT and Indicator
Where Acc is true accuracy,
not BIF accuracy
Example: Ultrasound
• Use live-animal ultrasound measurement
as an indicator of carcass merit
• Assume genetic correlation btw yearling
bull ultrasound and steer carcass is 0.70
• If a bull has an ultrasound EPD with
accuracy of 0.90, what is the accuracy of
selection for carcass?
Example: Ultrasound
If the 0.90 accuracy is “true” accuracy, then
the “true” accuracy of selection is:
0.90 x 0.70 = 0.63
But “BIF” accuracy would be quite different:
BIF Accuracy
(BIF) Accuracy of IMF
EPD
True Accuracy of
Selection for IMF
Genetic correlation
between IMF and MARB
True Accuracy of
Selection for Marbling
(BIF) Accuracy of
Selection for Marbling
0.90
0.50
0.20
0.995
0.866
0.600
0.70
0.70
0.70
0.696
0.606
0.420
0.282
0.205
0.092
Bottom Line
• BIF accuracy understates the true accuracy
of selection
• Selection on indicator traits only is
significantly less accurate unless the
genetic correlation is very high
Progeny Test
• How much additional accuracy is added by
a progeny test, assuming the animal is
phenotyped?
– Depends on the heritability of the trait
– Depends on the number of progeny
– Depends (a little) on how much pedigree
information is available on the animal
Progeny Test Example
• BIF accuracy values, calculated from
Selection Index
• Individual record on animal
• 10 or 200 Paternal half-sibs
• 0, 5, 50 or 500 progeny
• Heritability of 0.05, 0.25 or 0.45
Progeny Test Example
PHS
10
10
10
200
200
200
h2
.05
.25
.45
.05
.25
.45
No
Prog
.04
.17
.29
.11
.22
.32
BIF Accuracy
5
50
Prog
Prog
.07
.24
.25
.54
.38
.65
.13
.28
.29
.55
.44
.66
500
Prog
.63
.83
.88
.64
.83
.88
Progeny Test Example
• Progeny data add accuracy, regardless of
heritability
• Pedigree adds relatively little accuracy
especially when significant numbers of
progeny are evaluated
Adding Molecular Information
• Van Eenennaam (2010) showed that if
molecular breeding values (MBV)
explained all of the genetic variation for a
trait, the increase in genetic gain from
MAS would be substantial
• No products currently on the market have
been shown to explain a majority of the
variation for a quantitative trait
Adding Molecular Information
• MacNeil et al. (2010) compared the use of
ultrasound measurements and MBV for
prediction of carcass marbling BV in
Angus cattle
• Genetic correlation between ultrasound
IMF and carcass marbling was 0.56, and
0.38 between MBV and carcass marbling
Adding Molecular Information
• Despite the lower genetic correlation, the
authors expected greater genetic gain from
selection on MBV than on ultrasound,
because MBV are fully heritable
• Ultrasound values on sibs provide
significantly increased genetic gain, but
MBV on sibs does not
Adding Molecular Information
• Authors concluded that a very limited
amount of progeny carcass data would be
more informative than either MBV or
ultrasound
• MBV selection is more useful for traits that
are difficult to measure, or that are
measured later in life
Summary
• Accuracy of indicator trait EPD does not
reflect accuracy of selection for ERT
• Progeny records add significant
information, but adds both expense and
time to evaluations
• Molecular breeding values could be a very
useful tool for selection, if the tests explain
larger portions of the variation in a trait.