close

Вход

Log in using OpenID

embedDownload
Implementing single step
GBLUP in pigs
Andreas Hofer
SUISAG
SABRE-TP
12.06.2014 , Zug
12.06.2014
1
Outline
! What is single step GBLUP?
! Plan of implementation by SUISAG
! Validation of genetic evaluations
! First results
! Discussion
12.06.2014
SABRE-TP, Zug
2
What is single step GBLUP?
! Traditional several steps:
§  Estimation of SNP-effects based on training data
§  Estimate genomic breeding values for genotyped
selection canidates (dGBV)
§  Blending with traditional BVs to obtain GoBV
! Single step:
§  Combine phenotypic, pedigree and SNP-data in one
single analysis to directly obtain GoBV
§  Several studies show same or improved accuracy
§  Simpler to implement in established process of BV
estimation
12.06.2014
SABRE-TP, Zug
3
BLUP vs ssGBLUP
(Legarra,Christensen,Aguilar,Misztal: Single step, a general
approach for genomic selection, Livest Sci 2014)
! MME BLUP
! MME ssGBLUP
12.06.2014
SABRE-TP, Zug
4
G-1 – A22-1
! Sveral approaches to construct G (Van Raden, JDS 2008)
! G can be singular à Gα = (1-α)G+αA22 (usually 0.05≤ α≤0.2)
§  α can be viewed as proportion of genetic variance not
explained by the markers.
! G should be „on the same base“ as A
§  Base population allele freq. are not known à tr(G)=tr(A22)
§  Genotyped animals are not a random sample of base
animals (genetic trend, variance) à Gn = a + bG
a and b = function of tr(G), tr(A22), mean of G and A22
! Scale contributions à τG-1 - ωA22-1
(usually τ≤1, ω≤1)
! Combined: τ[a +b((1-α)G + αA22)]-1 - ωA22-1
12.06.2014
SABRE-TP, Zug
5
From traditional BLUP to
Single Step GBLUP
Marker
Marker
Imputing
Pedigree
Pedigree
Data base
Performance
Performance
BVs,
Computer Programs
accuracies
Genomic relationships
Pedigree relationships
BVs
byby
ssGBLUP
BVs
BLUP
Approx. accuracies
12.06.2014
SABRE-TP, Zug
6
Software
! Handling SNP-data: ?? so far own programs
! Imputing: Fimpute (Sargolzaei, licence fee)
! G-1 – A22-1: HGinv (Stranden, MTT, with Mix99)
§  Various options available: τ[a +b((1-α)G + αA22)]-1 - ωA22-1
! Solve MME: MiX99 (Lidauer, MTT, licence fee)
§  Variance components not yet implemented for ssGBLUP
§  But estimates are often very similar à little impact ??
! Approx. accuracies: ??
§  Modify own programs to add genomic contributions
according to Misztal et al. 2013 JDS.
12.06.2014
SABRE-TP, Zug
7
Validation of predicting ability of
estimated BVs
time
BLUP w‘out SNP
12.06.2014
All
available
performanc
e data
Pedigree
BVr =pedigree index
Benchmark BLUP w’out SNP
SNP-Daten
Validation
animals with
onw or prog.
performance
excl. here
Performanc
e data
available at
time x
Pedigree
x
Pedigree
Performanc
e data
available at
time x
ssGBLUP
Validation
animals with
onw or prog.
performance
excl. here
GoBVr
SABRE-TP, Zug
-  BV of validation animals
-  phenotypes of valid. animals
corrected for est. fixed effects
8
First results: Data sets
! Single trait analysis for litter size (NBA)
! Split data set by birth date of litter at 1.10.2011
! Validation animals have genotypes (60K) and own or
prog. performance only ≥ 1.10.2011
§  40 boars:
∅ each with 116 daughters with 2.9 litters/daugther
§  141 sows:
∅ each with 5 own litters, 81 also with daughters
12.06.2014
SABRE-TP, Zug
9
Accuracy (r2) of EBV of validation
animals in reduced and complete data
Reduced data set (4)
Complete data set (1-4)
(Pedigree Index)
Male
100
100
80
80
60
60
40
40
20
20
Count
Count
Male
Female
100
Female
100
80
80
60
60
40
40
20
20
0.2
0.4
0.6
0.8
0.2
Accuracy of BLUP (r2)
12.06.2014
0.4
0.6
0.8
Accuracy of BLUP (r2)
SABRE-TP, Zug
10
Gn vs A22 after exclusion of outliers
! Outliers =
§  High/low gii
§  Off-diagonals:
|aij-gij|>0.4
! 5 Pairs of duplicates
! Total 59 animals
eliminated
! Correlations A-G
§  Diagonals = -0.09 ??
§  Off-diagonals = 0.64
12.06.2014
SABRE-TP, Zug
11
Single trait runs
1
2ab
3abτω
4abτω
5old
6cut
BYear ped
1994
1994
1994
1994
1988
1996
BYear SNP
1996
1996
1996
1996
1992
2000
Phen from
2000
2000
2000
2000
1996
2004
α
0.1
0.1
0.1
0.1
0.1
0.1
a/b
0/1
τ/ω
1/1
12.06.2014
0.07/0.94 0.07/0.94 0.07/0.94 0.08/0.93 0.07/0.94
1/1
1.5/0.6
SABRE-TP, Zug
1/0.7
1/0.7
1/0.7
12
Reference animals in runs 1-4
! 533 males
Reference animals red data set (4) (N= 1418 )
r2â,a = 0.69
(only 254 with r2â,a >0.8)
Male
400
300
200
! 885 females
= 0.47
Count
r2â,a
100
0
Female
400
300
Ø Limited information from
SNPs, because gain in
accuracy for animal i =
f(Σ (gij-aij)2* r2â,aj ) (Misztal et al.
200
100
0
0.2
0.4
0.6
0.8
1.0
Accuracy of BLUP (r2)
2013)
12.06.2014
SABRE-TP, Zug
13
Validation criteria = forward
prediction of BV or phenotypes
! Breeding values of validation animals (40m,141f)
§  Relative increase of correlation with BVc (complete
data, no SNP)
r(GoBVr, BVc) / r(BVr, BVc)
§  Regression of BVc on GoBVr or BVr should be 1
BVc = b0 + b1 x GoBVr + e vs
BVc = b0 + b1 x BVr
+e
! Same for phenotypes corrected for fixed effects
(with complete data, no SNP) averaged per sow
§  r(GoBVr, y_cor) / r(BVr, y_cor)
§  y_cor = b0 + b1 x GoBVr + e vs
y_cor = b0 + b1 x BVr
12.06.2014
+e
SABRE-TP, Zug
14
Forward predicition of BVs
1
2ab
3abτω
4abτω
5old
6cut
BY2010-2004 c
1.29
1.29
1.29
1.29
1.38
1.18
BY2010-2004 r
1.18
1.18
1.18
1.18
1.26
1.09
BY2010-2004 rS
0.80
1.14
1.02
1.09
1.17
1.01
r(BVr, BVc) m
r(GoBVr,BVc) m
0.61
0.72
0.61
0.72
0.61
0.72
0.61
0.72
0.62
0.72
0.60
0.72
rel. increase
1.17
1.17
1.18
1.17
1.16
1.20
b1(BVr) m
0.86
0.86
0.86
0.86
0.88
0.85
b1(GoBVr) m
0.90
0.88
1.14
1.04
1.03
1.05
r(BVr, BVc) f
0.65
0.65
0.65
0.65
0.66
0.65
r(GoBVr,BVc) f
0.62
0.62
0.67
0.68
0.67
0.66
rel. increase
0.95
0.96
1.04
1.04
1.03
1.02
b1(BVr) f
0.93
0.93
0.93
0.93
0.95
0.94
b1(GoBVr) f
0.65
0.67
1.03
0.93
0.93
0.94
12.06.2014
SABRE-TP, Zug
15
Forward predicition of averaged
phenotypes of 141 valid. females
1
2ab
3abτω
4abτω
5old
6cut
r(GoBVr, y_cor)
0.21
0.28
0.21
0.25
0.21
0.25
0.21
0.26
0.22
0.26
0.22
0.25
rel. Increase
1.34
1.21
1.22
1.22
1.19
1.13
b1(BVr)
0.84
0.84
0.84
0.84
0.89
0.90
b1(GoBVr)
0.84
0.77
1.09
0.98
1.01
1.01
r(BVr, y_cor)
! Larger increase of accuracy if sire genotyped (not shown)
! Conclusion: 4 seems to provide „best“ predictions
§  Interaction among parameters α,a/b,τ/ω ?
§  Repeatable for other validation animals?
§  Best option for litter size also best for other traits?
Ø additonal validation runs needed!
12.06.2014
SABRE-TP, Zug
16
So far only moderate improv. of
predictions (plots for run 4)
Validation boars
6
4
2
-2
0
averaged corrected phenotypes
3
2
1
0
-1
EBVcompl w'out SNP
4
Validation phenotypes
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
EBVred w'out(red)/with(blue) SNP
12.06.2014
-1
0
1
2
3
4
5
EBVred w'out(red)/with(blue) SNP
SABRE-TP, Zug
17
Discussion
! Method of validation
§  Aim of genetic evaluation is prediction à forward
predicition is method of choice
§  Used BVc or y_cor as surrogate for the unknown true
breeding value of validation animals
-  Others use DYDs or deregressed proofs ?
-  The variable used should be highly correlated to true breeding value
and independent of prediction method (at least not favouring SNP)
! Lack of correlation bw diagonals of A and G
§  Diag(A) reflect inbreeding à pedigree depth?
-  Adding 2 years of pedigree in relation to SNP-animals did not help
§  Diag(G) reflect homozygosity, but E(G) = A
§  What could be potential problems?
12.06.2014
SABRE-TP, Zug
18
Diagonal elements of A and G vs
birth year (run 5)
1.12
600
t 1.1
n
e 1.08
m
e
l 1.06
e
l
a 1.04
n
o
ga
i 1.02
d
e
ga 1
r
e
v 0.98
A
0.96
0.94
12.06.2014
500
400
300
diag(A)
200
diag(G)
100
6
9
9
1
7
9
9
1
8
9
9
1
9
9
9
1
0
0
0
2
1
0
0
2
2 3 4
0 0 0
0 0 0
2 2 2
Birth year
5
0
0
2
6
0
0
2
SABRE-TP, Zug
7
0
0
2
8
0
0
2
9
0
0
2
0
1
0
2
1
1
0
2
N
0
19
PCA SNPs before Imputation
(only SNPs with no missing genotypes, run 6)
12.06.2014
SABRE-TP, Zug
20
PCA SNPs after Imputation
(every 2.5nd SNP, run 6)
12.06.2014
SABRE-TP, Zug
21
3
2
1
0
-2
-1
Ped_P$ZwPoSNP
4
BVr vs GoBVr (run 4)
-2
-1
0
1
2
3
4
Ped_P$ZwPmSNP
12.06.2014
SABRE-TP, Zug
22
GoBVr or BVr vs BVc for 40 boars
(run 4)
0
1
2
3
BVr:
r = 0.61, b1= 0.86
GoBVr:
r = 0.72, b1= 1.04
-1
EBVcompl w'out SNP
4
Validation boars
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
EBVred w'out(red)/with(blue) SNP
12.06.2014
SABRE-TP, Zug
23
GoBVr or BVr vs BVc for 141
females (run 4)
0
1
2
3
BVr:
r = 0.65, b1= 0.93
GoBVr:
r = 0.68, b1= 0.93
-1
EBVcompl w'out SNP
4
Validation sows
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
EBVred w'out(red)/with(blue) SNP
12.06.2014
SABRE-TP, Zug
24
GoBVr or BVr vs average of y_cor
for 141 females (run 4)
0
2
4
6
BVr:
r = 0.21, b1= 0.84
GoBVr:
r = 0.26, b1= 0.98
-2
averaged corrected phenotypes
Validation phenotypes
-1
0
1
2
3
4
5
EBVred w'out(red)/with(blue) SNP
12.06.2014
SABRE-TP, Zug
25
1/--pages
Пожаловаться на содержимое документа