![Page 1: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/1.jpg)
Genomic Selection in the era of Genome sequencing
![Page 2: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/2.jpg)
Course overview
• Day 1– Linkage disequilibrium in animal and plant genomes
• Day 2– Genome wide association studies
• Day 3 – Genomic selection
• Day 4 – Genomic selection
• Day 5– Imputation and whole genome sequencing for genomic
selection
![Page 3: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/3.jpg)
Genomic selection
• Introduction
• Genomic selection with Least Squares and BLUP
• Introduction to Bayesian methods
• Genomic selection with Bayesian methods
• Comparison of accuracy of methods
![Page 4: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/4.jpg)
Genomic selection
• Problem marker assisted selection is only a proportion of genetic variance is tracked with markers
– Eg. 10 QTL << 5% of the genetic variance
• Alternative is to trace all segments of the genome with markers
– Divide genome into chromosome segments based on marker intervals?
– Capture all QTL = all genetic variance
![Page 5: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/5.jpg)
Genomic selection
chromosomeM M M M M M M M M M M
![Page 6: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/6.jpg)
Genomic selection
chromosome
chromosome segment i
M M M M M M M M M M M
![Page 7: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/7.jpg)
Genomic selection
chromosome
chromosome segment i
M M M M M M M M M M M
chromosome segment effects gi
1_1 2_1
1_2 2_2
![Page 8: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/8.jpg)
Genomic selection
• Predict genomic breeding values as sum of effects over all segments
∑∧
=p
i
ii gXGEBV
![Page 9: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/9.jpg)
Genomic selection
• Predict genomic breeding values as sum of effects over all segments
∑∧
=p
i
ii gXGEBV
Number of chromosome segments
![Page 10: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/10.jpg)
Genomic selection
• Genomic selection can be implemented
–with marker haplotypes within chromosome segments
∑∧
=p
i
ii gXGEBV
1_1 0.3
1_2 0.0
2_1 -0.2
2_2 -0.1
![Page 11: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/11.jpg)
Genomic selection
• Genomic selection can be implemented
–with marker haplotypes within chromosome segments
–with single markers ∑∧
=p
i
ii gXGEBV
∑∧
=p
i
ii gXGEBV
2 -0.5
![Page 12: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/12.jpg)
Genomic selection
• Genomic selection exploits linkage disequilibrium
– Assumption is that effect of haplotypes or markers within chromosome segments picking up QTL and will have same effect
across the whole population
• Possible within dense marker maps now available 1_1 0.3
1_2 0.0
2_1 -0.2
2_2 -0.1
![Page 13: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/13.jpg)
Genomic selection
• Genomic selection avoids bias in estimation of effects due to multiple testing, as all effects fitted simultaneously
![Page 14: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/14.jpg)
Genomic selection
![Page 15: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/15.jpg)
Genomic selection
• First step is to predict the chromosome segment effects in a reference population
• Number of effects >>> than number of records
• Eg. 10 000 intervals * 4 haplotypes = 40 000 haplotype effects
• From ~ 2000 records?
• Need methods that can deal with this
![Page 16: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/16.jpg)
Genomic selection
• Introduction
• Genomic selection with Least Squares and BLUP
• Introduction to Bayesian methods
• Genomic selection with Bayesian methods
• Comparison of accuracy of methods
![Page 17: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/17.jpg)
Least squares Genomic selection
• Two step procedure– Test each chromosome segment for presence of QTL
(fitting haplotypes within segment), take significant effects
– Fit the significant effects simultaneously in multiple regression
– Predict GEBVs
• Identical to Marker assisted selection with multiple markers
• Problems remain– Do not capture all QTL– Over-estimation of haplotype effects due to setting of
significance threshold
![Page 18: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/18.jpg)
Genomic selection with BLUP
• BLUP = best linear unbiased prediction
• Model:
• In BLUP we assume variance of haplotype effects across all segments is equal, eg E(g) ~ N(0,σg
2), where g = [g1g2g3..gp]
egX1y iin ++= ∑=
p
i 1
µ
![Page 19: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/19.jpg)
Genomic selection with BLUP
• BLUP assumes normal distribution of SNP/haplotype effects
-4 -2 0 2 4 6
0.0
0.1
0.2
0.3
0.4
De
nsity
![Page 20: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/20.jpg)
Genomic selection with BLUP
• BLUP = best linear unbiased prediction
• Then we can estimate segment effects as:
• λ=σe2 / σg
2
+=
−
∧
∧
yX'
y'1
IXX'X'1
X'1'11
g
n
1
n
nnn
λ
µ
![Page 21: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/21.jpg)
Genomic selection with BLUP
• Example
• A “simulated” data set
• Single chromosome, with 10 markers
• Phenotypes “simulated”
– overall mean of 1
– an effect for SNP 1 of 2 allele of 1
– normally distributed error term with mean 0 and variance 1.
![Page 22: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/22.jpg)
Genomic selection with BLUP
• Example
• 10 SNPs
• Only 5 phenotypic records.
X
Animal Y 1 2 3 4 5 6 7 8 9 10
1 0.19 0 0 0 0 0 0 1 2 0 2
2 1.23 1 0 0 1 1 1 2 1 0 1
3 0.86 1 0 0 1 0 0 1 1 1 1
4 1.23 1 1 1 1 0 1 2 1 1 1
5 0.45 0 1 1 1 1 1 2 1 0 1
![Page 23: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/23.jpg)
Genomic selection with BLUP
• Example
• Assume value of 1 for λ
• 1n = [1 1 1 1 1]
+=
−
∧
∧
yX'
y'1
IXX'X'1
X'1'11
g
n
1
n
nnn
λ
µ
![Page 24: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/24.jpg)
Genomic selection with BLUP
• Example
Mean 0.47
SNP1 0.29
SNP2 -0.05
SNP3 -0.05
SNP4 0.08
SNP5 -0.02
SNP6 0.13
SNP7 0.13
SNP8 -0.08
SNP9 0.11
SNP10 -0.08
![Page 25: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/25.jpg)
Genomic selection with BLUP
• Now we want to predict GEBV for a group of young animals without phenotypes.
• We have the g_hat, and we can get X from their haplotypes (after genotyping)…………
∧
= gXGEBV
Progeny X
1 1 1 1 1 1 1 2 1 0 1
2 1 0 0 1 1 1 2 1 0 1
3 1 0 0 1 1 1 2 1 0 1
4 1 0 0 1 1 1 2 1 0 1
5 0 0 0 0 0 0 1 2 0 2
![Page 26: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/26.jpg)
Genomic selection with BLUP
• GEBV∧
= gXGEBV
X GEBV
∧
g
1 1 1 1 1 1 2 1 0 1 0.29 0.47
1 0 0 1 1 1 2 1 0 1 -0.05 0.58
1 0 0 1 1 1 2 1 0 1 -0.05 0.58
1 0 0 1 1 1 2 1 0 1 0.08 0.58
0 0 0 0 0 0 1 2 0 2 -0.02 -0.20
0.13
0.13
-0.08
0.11
-0.08
![Page 27: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/27.jpg)
Genomic selection with BLUP
• Where do we get σg2 from?
• Can estimate total additive genetic variance and divide by number of segments, eg σg
2 = σa2 /p
• If using single markers take account of heterozygosity
• Ridge regression (Bayesian approach)• Cross validation
∑=
−=p
i
iiag qq1
22 )1(2/σσ
![Page 28: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/28.jpg)
Genomic selection with BLUP
• An equivalent model
• If there are many QTLs whose effects are normally distributed with constant variance,
• Then genomic selection equivalent to replacing the expected relationship matrix with the realised or genomic relationship matrix (G) estimated from
DNA markers in normal BLUP equations.
– Gij = proportion of genome that is IBD between animals i and j
![Page 29: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/29.jpg)
Genomic selection with BLUP
• An equivalent model
• Rescale X to account for allele frequencies
–wij = xij – 2pj
• Then breeding values are
– v = Wg ( )
• And
• Then
∧
= gXGEBV
∑=
−=p
j
jj pp1
)1(2/WW'G
2)( aV σGv =
![Page 30: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/30.jpg)
Genomic selection with BLUP
• An equivalent model
eZv1y n ++= µ
+=
−
−∧
∧
yZ'
y'1
GZZ'Z'1
Ζ'1'11
v
n
1
1
n
νnn
2
2
a
e
σ
σµ
![Page 31: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/31.jpg)
Genomic selection with BLUP
• An equivalent model
–Model 1.
–Model 2.
egX1y iin ++= ∑=
p
i 1
µ∧
= gXGEBV
+=
−
∧
∧
yX'
y'1
IXX'X'1
X'1'11
g
n
1
n
nnn
2
2
g
e
σ
σµ
![Page 32: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/32.jpg)
Genomic selection with BLUP
• An equivalent model
–Model 1.
–Model 2.
eZv1y n ++= µ
egX1y iin ++= ∑=
p
i 1
µ∧
= gXGEBV
+=
−
−∧
∧
yZ'
y'1
GZZ'Z'1
Ζ'1'11
v
n
1
1
n
νnn
2
2
v
e
σ
σµ
+=
−
∧
∧
yX'
y'1
IXX'X'1
X'1'11
g
n
1
n
nnn
2
2
g
e
σ
σµ
![Page 33: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/33.jpg)
Holstein reference n = 781
Jersey reference n = 287
Holstein validation n = 400
Jersey validation n = 77
![Page 34: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/34.jpg)
Genomic selection with BLUP
• An equivalent model
• Why use model 2.– If number of markers >>> large than number
of animals, more computationally efficient
– Can be integrated into national evaluations more readily?
– Calculate accuracy of GEBV from inverse coefficient matrix
![Page 35: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/35.jpg)
Genomic selection
![Page 36: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/36.jpg)
Genomic selection
![Page 37: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/37.jpg)
Genomic selection
• Introduction
• Genomic selection with Least Squares and BLUP
• Introduction to Bayesian methods
• Genomic selection with Bayesian methods
• Comparison of accuracy of methods
![Page 38: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/38.jpg)
Bayesian methods
• BLUP assumes normally distributed QTL effects
• Does not match prior knowledge of distributions of QTL effects for some traits
• Use Bayesian approaches to incorporate prior knowledge 0.000 0.005 0.010 0.015 0.020
05
00
00
10
00
00
150
00
02
00
00
02
50
00
03
00
00
0
STD Effects
de
nsity
PYFYMYPCFC
![Page 39: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/39.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
![Page 40: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/40.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Probability of
parameters x given
the data y (posterior)
![Page 41: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/41.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Probability of
parameters x given
the data y (posterior)
Is proportional to
![Page 42: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/42.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Probability of
parameters x given
the data y (posterior)
Is proportional to Probability of
data y given the
x (likelihood of
data)
![Page 43: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/43.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Probability of
parameters x given
the data y (posterior)
Is proportional to Probability of
data y given the
x (likelihood of
data)
Prior
probability
of x
![Page 44: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/44.jpg)
Bayesian methods
• Consider an experiment where we measure height of 10 people to estimate average height
• We want to use prior knowledge from many previous studies that average height is 174cm with standard error 5cm
y=average height + e
![Page 45: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/45.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Prior probability of x (average height)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
Den
sity
![Page 46: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/46.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Prior probability of x (average height)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
Den
sity
5.
178
=
=
es
x
From the data……
![Page 47: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/47.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
Prior probability of x (average height)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
Den
sity
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
L(y
|x)
Likelihood of data (y) given
height x, most likely x = 178cm
![Page 48: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/48.jpg)
Bayesian methods• Bayes theorem
)()|()|( xPxyPyxP ∝
L(y|x) P(x)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
De
nsity
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
L(y
|x)
0
0.001
0.002
0.003
0.004
0.005
0.006
160 165 170 175 180 185 190
Height
P(H
eig
ht|y)
P(x|y) mean = 176cm
![Page 49: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/49.jpg)
Bayesian methods• Bayes theorem
• Less certainty about prior information? Use less informative (flat) prior
)()|()|( xPxyPyxP ∝
L(y|x) P(x)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
L(y
|x)
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
160 165 170 175 180 185 190
Height
De
nsity
![Page 50: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/50.jpg)
Bayesian methods• Bayes theorem
• Less certainty about prior information? Use less informative (flat) prior
)()|()|( xPxyPyxP ∝
L(y|x) P(x)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
L(y
|x)
P(x|y) mean = 178cm
0
0.001
0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
160 165 170 175 180 185 190
Height
De
nsity
0
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
0.0007
160 165 170 175 180 185 190
Height
P(H
eig
ht|y)
![Page 51: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/51.jpg)
Bayesian methods• Bayes theorem
• More certainty about prior information? Use more informative prior
)()|()|( xPxyPyxP ∝
L(y|x) P(x)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
L(y
|x)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
160 165 170 175 180 185 190
Height
P(h
eig
ht)
![Page 52: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/52.jpg)
Bayesian methods• Bayes theorem
• More certainty about prior information? Use more informative prior
)()|()|( xPxyPyxP ∝
L(y|x) P(x)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
160 165 170 175 180 185 190
Height
L(y
|x)
P(x|y) mean = 174.5cm
0
0.005
0.01
0.015
0.02
0.025
160 165 170 175 180 185 190
Height
P(H
eig
ht|
y)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
160 165 170 175 180 185 190
Height
P(h
eig
ht)
![Page 53: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/53.jpg)
Genomic selection
• Introduction
• Genomic selection with Least Squares and BLUP
• Introduction to Bayesian methods
• Genomic selection with Bayesian methods
• Comparison of accuracy of methods
![Page 54: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/54.jpg)
Genomic selection
• For some traits prior knowledge suggests t-distribution of effects
• How to incorporate this into our predictions?
0.000 0.005 0.010 0.015 0.020
05
00
00
10
00
00
15
00
00
20
00
00
25
00
00
30
00
00
STD Effects
de
nsity
PYFYMYPCFC
![Page 55: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/55.jpg)
Genomic selection
• For some traits prior knowledge suggests t-distribution of effects
• How to incorporate this into our predictions?
-10 -5 0 5 10
0.0
0.1
0.2
0.3
0.4
x
Den
sity
![Page 56: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/56.jpg)
Genomic selection
• The t distributioncan be presented as a two level hierarchical model
• Allow different variances between chromosome segments
• Assume a distribution of these variances
• Computationally easier to deal with than original form -10 -5 0 5 10
0.0
0.1
0.2
0.3
0.4
x
Den
sity
![Page 57: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/57.jpg)
Genomic selection
chromosome
chromosome segment 5
M M M M M M M M M M M
chromosome segment effects g5
1_1 2_1
1_2 2_2
σg52
![Page 58: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/58.jpg)
Genomic selection
chromosome
chromosome segment 5, 10
M M M M M M M M M M M
chromosome segment effects g5, g10
1_1 2_1
1_2 2_2
σg52
1_1 2_1
1_2 2_2
σg102
![Page 59: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/59.jpg)
Genomic selection
chromosome
chromosome segment 5, 10
M M M M M M M M M M M
chromosome segment effects g5, g10
1_1 2_1
1_2 2_2
σg52
1_1 2_1
1_2 2_2
σg102
=
![Page 60: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/60.jpg)
Genomic selection
chromosome
chromosome segment 5, 10
M M M M M M M M M M M
chromosome segment effects g5, g10
1_1 2_1
1_2 2_2
σg52
1_1 2_1
1_2 2_2
σg102
=g1,g2,g3,g4,g5…….. ~ N(0, σg2)
![Page 61: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/61.jpg)
Genomic selection
chromosome
chromosome segment 5, 10
M M M M M M M M M M M
chromosome segment effects g5, g10
1_1 2_1
1_2 2_2
σg52
1_1 2_1
1_2 2_2
σg102
≠
![Page 62: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/62.jpg)
Genomic selection
chromosome
chromosome segment 5, 10
M M M M M M M M M M M
chromosome segment effects g5, g10
1_1 2_1
1_2 2_2
σg52
1_1 2_1
1_2 2_2
σg102
≠g5~ N(0, σg52), g10~ N(0, σg10
2)
![Page 63: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/63.jpg)
Bayesian methods
• Now lets allow different variances of chromosome segment effects
+
+
=
−
∧
∧
∧
yX
yX
y
σ
σ
σ
σ
p
n
gp
e
g
e
p
'
.
'
'1
.
1
2
2
2
1
2
1
1
pp1pnp
p111n1
pn1nnn
IX'X.X'X'1X
....
X'X.IX'X'1X
X'1.X'1'11
g
g
µ
![Page 64: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/64.jpg)
0.05 0.10 0.15 0.20 0.25
010
20
30
40
50
X
Y
-10 -5 0 5 10
0.0
0.1
0.2
0.3
0.4
x
De
nsity
Distribution of σgj2 --���� Distribution of gj
![Page 65: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/65.jpg)
Bayesian methods
• Now lets allow different variances of chromosome segment effects
• Need two levels of models
– Data
– Variances of chromosome segment effects
)()|()|( 222
gigiiigi PgPgP σσσ ∝
),(),|()|,( µµµ ggg PyPyP ∝
![Page 66: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/66.jpg)
Bayesian methods
• Now lets allow different variances of chromosome segment effects
• Data
),(),|()|,( µµµ ggg PyPyP ∝
+
+
=
−
∧
∧
∧
yX
yX
y
σ
σ
σ
σ
p
n
gp
e
g
e
p
'
.
'
'1
.
1
2
2
2
1
2
1
1
pp1pnp
p111n1
pn1nnn
IX'X.X'X'1X
....
X'X.IX'X'1X
X'1.X'1'11
g
g
µ
![Page 67: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/67.jpg)
Bayesian methods
• Variances of chromosome segments
• Note that these variance components are not the parameters of interest
• However they are useful intermediates to arrive at better inferences for the gi
• Amount of shrinkage of effects varies between segments
)()|()|( 222
gigiiigi PgPgP σσσ ∝
![Page 68: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/68.jpg)
Bayesian methods
• Variances of chromosome segments
• Prior?
– Inverted chi square convenient for variances
)()|()|( 222
gigiiigi PgPgP σσσ ∝
![Page 69: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/69.jpg)
Bayesian methods
• Prior?– Inverted chi square convenient for variances– An inverted chi square with v degrees of freedom
and scaled by S2, eg.
– Describes a distribution with • mean
• variance
– Larger v, more informative prior = more belief about variance
22 / vS χ
)2/(2 −vvS
)4()2(
22
42
−− vv
Sv
![Page 70: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/70.jpg)
Bayesian methods
v=2
![Page 71: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/71.jpg)
Bayesian methods
v=2
v=20
![Page 72: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/72.jpg)
Bayesian methods
• Variances of chromosome segments
• Prior?
• We can choose v and S2 so that the prior reflects our knowledge that there are many QTL of small effect and few of large effect
)()|()|( 222
gigiiigi PgPgP σσσ ∝
22 / vS χ
![Page 73: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/73.jpg)
0.05 0.10 0.15 0.20 0.25
010
20
30
40
50
X
Y
-10 -5 0 5 10
0.0
0.1
0.2
0.3
0.4
x
De
nsity
Distribution of σgj2 --���� Distribution of gj
![Page 74: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/74.jpg)
Bayesian methods
0
0.1
0.2
0.3
0.4
0.5
0 0.2 0.4 0.6 0.8 1
Size of QTL (phenotypic standard deviations)
Pro
po
rtio
n o
f Q
TL
2
)002.0,012.4(
−χ
E(σgi2)=S/(v-2)
V(σgi2)/ [E(σgi2)]2=2/(v-4)
![Page 75: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/75.jpg)
Bayesian methods
• Variances of chromosome segments
• Posterior?
– An advantage of choosing the inverse chi-square distribution for the prior is that the posterior will also be an inverse chi-square distribution
• Degrees of freedom = prior + data
• Scaling factor = sums of squares prior (S2) + sums of squares from data
)()|()|( 222
gigigi PPP σσσ ii gg ∝
![Page 76: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/76.jpg)
Bayesian methods
• Variances of chromosome segments
• Posterior?
– ni = number of haplotype effects
)()|()|( 222
gigigi PPP σσσ ii gg ∝
2
),( 2
−
++ ii g'gSnv i
χ
![Page 77: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/77.jpg)
Bayesian methods
• Variances of chromosome segments
• Posterior?
• But posterior cannot be estimated directly, dependent on gi!!
2
)002.0,012.4(
−++ ii g'ginχ
)()|()|( 222
gigigi PPP σσσ ii gg ∝
![Page 78: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/78.jpg)
Bayesian methods
• Solution is to use Gibbs sampling
– Draw samples from the posterior distributions of parameters conditional on all other effects
– The average of these samples can be used as the estimates of the parameters
![Page 79: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/79.jpg)
Bayesian methods
• Gibbs sampling scheme
–Parameters to estimate and their posteriors
–P(σgi2|gi)
–P(σe2|e)
–P(µ|y,e,g, σe2)
–P(gij|y,µ,g≠ij,σgi2,σe
2)
2
)002.0,012.4(
−++ ii g'ginχ
2
),2(
−− ee'nχ
( )
− n
nN e /,
1 2σXg1y1 '
n
'
n0
0 .0 5
0 .1
0 .1 5
0 .2
0 .2 5
0 .3
0 .3 5
0 .4
0 .4 5
1 6 0 1 6 5 1 7 0 1 7 5 1 8 0 1 8 5 1 9 0
H e ig h t
P(h
eig
ht)
( )
+
+
−− = 222
22//,
/gieije
gie
N σσσσσ
µX'X
XX
1XXgXyXij
ij
'
ij
n
'
ij0)(ij
'
ij
'
ij
![Page 80: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/80.jpg)
Bayesian methods
• Gibbs sampling scheme
–Parameters to estimate and their posteriors
–P(σgi2|gi)
–P(σe2|e)
–P(µ|y,e,g, σe2)
–P(gij|y,µ,g≠ij,σgi2,σe
2)
2
)002.0,012.4(
−++ ii g'ginχ
2
),2(
−− ee'nχ
( )
− n
nN e /,
1 2σXg1y1 '
n
'
n0
0 .0 5
0 .1
0 .1 5
0 .2
0 .2 5
0 .3
0 .3 5
0 .4
0 .4 5
1 6 0 1 6 5 1 7 0 1 7 5 1 8 0 1 8 5 1 9 0
H e ig h t
P(h
eig
ht)
( )
+
+
−− = 222
22//,
/gieije
gie
N σσσσσ
µX'X
XX
1XXgXyXij
ij
'
ij
n
'
ij0)(ij
'
ij
'
ij
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
-3 -2 -1 0 1 2 3
Effect
De
ns
ity
![Page 81: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/81.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
2
)002.0,012.4(
−++ ii g'ginχ
![Page 82: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/82.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
• σg12=0.95
2
)002.0,012.4(
−++ ii g'ginχ
![Page 83: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/83.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
–Step 3. Draw a sample from P(σe2|e)
First calculate the e as
µ'1n−−= Xgye
![Page 84: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/84.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
–Step 3. Draw a sample from P(σe2|e)
First calculate the e as
–Then sample…
µ'1n−−= Xgye
2
),2(
−− ee'nχ
![Page 85: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/85.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
–Step 3. Draw a sample from P(σe2|e)
First calculate the e as
–Then sample…
– σe2 = 0.5
µn1−−= Xgye
2
),2(
−− ee'nχ
![Page 86: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/86.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
–Step 3. Draw a sample from P(σe2|e)
–Step 4. Draw a sample from P(µ|y,g,σe2)
![Page 87: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/87.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
–Step 3. Draw a sample from P(σe2|e)
–Step 4. Draw a sample from P(µ|y,g,σe2)
– µ=-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
mean
den
sit
y
( )
− n
nN e /,
1 2σXg1y1 '
n
'
n
![Page 88: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/88.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, eg. g=0.01 and µ, eg µ=0.01
–Step 2. For each i, draw from P(σgi2|gi)
–Step 3. Draw a sample from P(σe2|e)
–Step 4. Draw a sample from P(µ|y,g,σe2)
–Step 5. For each gij, draw from P(gij|y,µ,g,σgi
2,σe2)
– g11 = 0.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
mean
de
ns
ity
![Page 89: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/89.jpg)
Bayesian methods
• The Gibbs chain
–Repeat steps 2-5 many times to build up samples from posterior distributions of the parameters
![Page 90: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/90.jpg)
Bayesian methods
• The Gibbs chain
–Repeat steps 2-5 many times to build up samples from posterior distributions of the parameters
–Finally, take estimates of parameters as average over many cycles
–Discard first ~ 100 cycles as dependent on starting values
![Page 91: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/91.jpg)
Bayesian methods
• Example– Consider a data set with three markers. The data
set was simulated as: – the effect of a 2 allele at the first marker is 3, the
effect of a 2 allele at the second marker is 0, and the effect of a 2 allele at the third marker was -2.
– the µ was 3 – σe
2 was 0.23. The data set was:
![Page 92: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/92.jpg)
Bayesian methods
• Example
Animal Phenotype Marker1 allele 1 Marker1 allele 2 Marker2 allele 1 Marker 2 allele 2 Marker3 allele 1 Marker 3 allele 2
1 9.68 2 2 2 1 1 1
2 5.69 2 2 2 2 2 2
3 2.29 1 2 2 2 2 2
4 3.42 1 1 2 1 1 1
5 5.92 2 1 1 1 1 1
6 2.82 2 1 2 1 2 2
7 5.07 2 2 2 1 2 2
8 8.92 2 2 2 2 1 1
9 2.4 1 1 2 2 1 2
10 9.01 2 2 2 2 1 1
11 4.24 1 2 1 2 2 1
12 6.35 2 2 1 1 1 2
13 8.92 2 2 1 2 1 1
14 -0.64 1 1 2 2 2 2
15 5.95 2 1 1 1 1 1
16 6.13 1 2 2 1 1 1
17 6.72 2 1 2 1 1 1
18 4.86 1 2 2 1 1 2
19 6.36 2 2 2 2 2 2
20 0.81 1 1 2 1 1 2
21 9.67 2 2 1 2 1 1
22 7.74 2 2 2 1 1 2
23 1.45 1 1 2 2 2 1
24 1.22 1 1 2 1 2 1
25 -0.52 1 1 2 2 2 2
![Page 93: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/93.jpg)
Bayesian methods
• Example– The Bayesian approach was applied, fitting
single marker effects
– X matrix
• Number of copies of two allele for each animal, eg. 2 1 0 for animal 1.
![Page 94: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/94.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
![Page 95: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/95.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
2
)002.0,012.4(
−++ ii g'ginχ
![Page 96: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/96.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
• σg12=0.002, σg2
2=0.06, σg32=0.009
2
)001.0002.0,1012.4(
−++χ
![Page 97: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/97.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
• σg12=0.002, σg2
2=0.06, σg32=0.009
–Step 3. Draw a sample from P(σe2|e)
µn1−−= Xgye
2
),2(
−− ee'nχ
![Page 98: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/98.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
• σg12=0.002, σg2
2=0.06, σg32=0.009
–Step 3. Draw a sample from P(σe2|e)
• σe2= 53.38
2
)031.812,23(
−χ
![Page 99: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/99.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
• σg12=0.002, σg2
2=0.06, σg32=0.009
–Step 3. Draw a sample from P(σe2|e)
• σe2= 53.38
–Step 4. Draw a sample from P(µ|y,g,σe2)
• µ=3.25
( )
− n
nN e /,
1 2σXg1y1 '
n
'
n
![Page 100: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/100.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
• σg12=0.002, σg2
2=0.06, σg32=0.009
–Step 3. Draw a sample from P(σe2|e)
• σe2= 53.38
–Step 4. Draw a sample from P(µ|y,g,σe2)
• µ=3.25
–Step 5. Draw a sample from P(gij|y,µ,g≠ij,σgi
2,σe2) ( )
+
+
−− = 222
22//,
/gieije
gie
N σσσσσ
µX'X
XX
1XXgXyXij
ij
'
ij
n
'
ij0)(ij
'
ij
'
ij
![Page 101: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/101.jpg)
Bayesian methods
• The Gibbs chain
–Step 1. Initialise value of g, µ
• g1=0.01, g2=0.01,g3=0.01, µ=0.1
–Step 2. For i=1,2,3, draw from P(σgi2|gi)
• σg12=0.002, σg2
2=0.06, σg32=0.009
–Step 3. Draw a sample from P(σe2|e)
• σe2= 53.38
–Step 4. Draw a sample from P(µ|y,g, σe2,e)
• µ=3.25
–Step 5. Draw a sample from P(gij|y,µ,g≠ij,σgi
2,σe2)
• g1=-0.02, g2=-0.81,g3=-0.005
![Page 102: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/102.jpg)
Bayesian methods
• Gibbs chain for 1000 cycles
– P(g1|y,µ,g≠1,σg12,σe
2)
![Page 103: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/103.jpg)
Bayesian methods
• Gibbs chain for 1000 cycles
– P(g1|y,µ,g≠1,σg12,σe
2)
“Burn in”
![Page 104: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/104.jpg)
Bayesian methods
• Gibbs chain for 1000 cycles
– P(g1|y,µ,g≠1,σg12,σe
2)
97.21 =∧
g
![Page 105: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/105.jpg)
Bayesian methods
• Gibbs chain for 1000 cycles
97.21 =∧
g 002.02 =∧
g 81.11 −=∧
g
![Page 106: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/106.jpg)
Bayesian methods
97.21 =∧
g 002.02 =∧
g 81.11 −=∧
g
Vector of SNP effects for calculating GEBV
![Page 107: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/107.jpg)
Bayesian methods
• Alternative priors for variance of segment haplotype/snp effects
–Meuwissen BayesA
–Xu (2003)
• Uninformative
–Te Braak (2006)
–Meuwissen BayesB
2
)002.0,012.4(
−χ
2
21/' −− aii gg χ
2
)002.0,012.4(
−++ ii g'ginχ
ασσ +−∝ 122 )()( gigip
2
)0,0(
−χ2
),1(
−gg'χ
![Page 108: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/108.jpg)
Bayesian methods
• Meuwissen BayesB– BayesA prior information is
many QTL with small effects and few with moderate effects
– But we have more prior knowledge than this – some chromosome segments will have no effect at all (contain no QTL)
• σgi2=0,gi =0
– How to sample from the posterior? -15 -10 -5 0 5 10 15 20
01
000
200
030
00
40
00
500
0
x
De
nsity
![Page 109: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/109.jpg)
Bayesian methods
• Meuwissen BayesB– If we sample σgi
2 from
–We will never sample 0, as the distribution has no mass at zero.
2
)002.0,012.4(
−++ ii g'ginχ
0 5 10 15 20 25
0.0
0.5
1.0
1.5
density.default(x = x)
N = 10000 Bandwidth = 0.04582
De
nsity
![Page 110: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/110.jpg)
Bayesian methods
• Meuwissen BayesB– If we sample σgi
2 from
–We will never sample 0 if gi’gi>0, as the distribution has no mass at zero.
–But if σgi2 >0, then sampling gi = 0 has
infinitesimal (basically zero) probability
2
)002.0,012.4(
−++ ii g'ginχ
![Page 111: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/111.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
We want to sample from this Can do it by sampling from these two
distributions
![Page 112: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/112.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
We want to sample from this P(gi|y,µ,g,σgi2,σe
2)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
mean
densit
y
![Page 113: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/113.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
??
Sample σgi2 without conditioning on gi
![Page 114: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/114.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Cannot be expressed as a known distribution = cannot use Gibbs for this bit
–Use a Metropolis Hastings algorithm
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
![Page 115: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/115.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Step1 Sample σg_new2,from prior(σg_new
2)
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
0.000 0.001 0.002 0.003 0.0040
10
00
200
03
00
04
00
0
density.default(x = xx)
De
nsi
ty
![Page 116: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/116.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Step1 Sample σg_new2,from prior(σg_new
2)
– σg_new2=0
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
0.000 0.001 0.002 0.003 0.0040
10
00
200
03
00
04
00
0
density.default(x = xx)
De
nsi
ty
![Page 117: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/117.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Step1 Sample σg_new2,from prior(σg_new
2)
– σg_new2=0.5
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
0.000 0.001 0.002 0.003 0.0040
10
00
200
03
00
04
00
0
density.default(x = xx)
De
nsi
ty
![Page 118: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/118.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Step 1 Sample σg_new2,from prior(σg_new
2)
–Step 2 Evaluate p(y*| σg_new2) (Likelihood)
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
*))*(*5.0(||2
1|(
2/12/1
2 yV'yV
y* 1−−= eLnginew
πσ ))2 2
eIσX'X(IV += ignewσ
![Page 119: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/119.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Step 1 Sample σg_new2,from prior(σg_new
2)
–Step 2 Evaluate p(y*| σg_new2) (Likelihood)
–Step 3 Replace σgi2 with σg_new
2 probability min[p(y*| σg_new
2)/ p(y*| σgi2):1]
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
![Page 120: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/120.jpg)
Bayesian methods
• Meuwissen BayesB
–Solution: sample σgi2,gi simultaneously from
the distribution:
–Step 1 Sample σg_new2,from prior(σg_new
2)
–Step 2 Evaluate p(y*| σg_new2) (Likelihood)
–Step 3 Replace σgi2 with σg_new
2 probability min[p(y*| σg_new
2)/ p(y*| σgi2):1]
–Step 4 Repeat ~ 100 cycles
*),|(*)|(*)|,( 222ygpypygp giigiigi σσσ ×=
![Page 121: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/121.jpg)
Genomic selection
• Introduction
• Genomic selection with Least Squares and BLUP
• Introduction to Bayesian methods
• Genomic selection with Bayesian methods
• Comparison of accuracy of methods
![Page 122: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/122.jpg)
Genomic selection
• Comparison of accuracy of methods (Meuwissen et al. 2001)– Genome of 1000 cM simulated, marker
spacing of 1 cM.
– Markers surrounding each 1-cM region combined into haplotypes.
– Due to finite population size (Ne = 100), marker haplotypes were in linkage disequilibrium with QTL between markers.
– Effects of haplotypes predicted in one generation of 2000 animals
– Breeding values for progeny of these animals predicted based on marker genotypes
![Page 123: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/123.jpg)
Genomic selection
• Comparison of accuracy of methods (Meuwissen et al. 2001)
rTBV;EBV + SE bTBV.EBV + SE
LS 0.318 ± 0.018 0.285 ± 0.024
BLUP 0.732 ± 0.030 0.896 ± 0.045
BayesA 0.798 0.827
BayesB 0.848 + 0.012 0.946 + 0.018
![Page 124: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/124.jpg)
Genomic selection
• Comparison of accuracy of methods (Meuwissen et al. 2001)– The least squares method does very poorly,
primarily because the haplotype effects are over-estimated.
![Page 125: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/125.jpg)
Genomic selection
• Comparison of accuracy of methods (Meuwissen et al. 2001)– The least squares method does very poorly,
primarily because the haplotype effects are over-estimated.
– Increased accuracy of the Bayesian approach because method sets many of the effects of the chromosome segments close to zero in BayesA, or zero in BayesB
![Page 126: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/126.jpg)
Genomic selection
• Comparison of accuracy of methods (Meuwissen et al. 2001)– The least squares method does very poorly,
primarily because the haplotype effects are over-estimated.
– Increased accuracy of the Bayesian approach because method sets many of the effects of the chromosome segments close to zero in BayesA, or zero in BayesB
– Also “shrinks” estimates of effects of other chromosome segments based on a prior distribution of QTL effects.
![Page 127: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/127.jpg)
Genomic selection
• Comparison of accuracy of methods (Meuwissen et al. 2001)– The least squares method does very poorly,
primarily because the haplotype effects are over-estimated.
– Increased accuracy of the Bayesian approach because method sets many of the effects of the chromosome segments close to zero in BayesA, or zero in BayesB
– Also “shrinks” estimates of effects of other chromosome segments based on a prior distribution of QTL effects.
– Accuracies were very high, as high as following progeny testing for example
![Page 128: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/128.jpg)
In real data
• 1500 Australian dairy
bulls
• genotyped for 56000
genome wide SNPs
• Phenotypes average
of daughters milk
production
![Page 129: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/129.jpg)
• Split data into two sub-populations
– Reference: Bulls born < 2003
– Validation: Bulls born >= 2003
In real data
![Page 130: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/130.jpg)
• Split data into two sub-populations
– Reference: Bulls born < 2003
– Validation: Bulls born >= 2003
• Accuracy
– Correlation of genomic breeding values with EBVs (which include daughter information) in validation set
In real data
![Page 131: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/131.jpg)
In real dataTable 3 MEBV- Correlation between predicted MEBV and ABV in the validation
data set (Bulls proven in years 2005, 2006, 2007)
Method Protein kg Fat kg Protein % Fat %
Bayes SSVS 0.55 0.51 0.68 0.73
Bayes A 0.53 0.48 0.66 0.70
BLUP 0.60 0.48 0.66 0.64
0.000 0.005 0.010 0.015 0.020
05
00
00
10
00
00
150
00
02
00
00
025
00
00
30
00
00
STD Effects
de
nsity
PYFYMYPCFC
BB
![Page 132: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/132.jpg)
Bayesian methods
97.21 =∧
g 002.02 =∧
g 81.11 −=∧
g
Vector of SNP effects for calculating GEBV
BLUP
![Page 133: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/133.jpg)
Genomic selection
• Yi and Xu 2008 (Genetics)
• Sample from inverse chi square distribution, but then sample shape (v) and scale (S2) of the distribution– Reflect absence of knowledge of distribution of QTL
effects?
– Prior on S2 is uniform, then posterior is gamma
– Prior on v of 1/v, not a conjugate prior = metropolis hastings
∑
=
p
j gj
vpvgammavyS
12
22 1
2,
2~,,,,|
σσβµ
![Page 134: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/134.jpg)
Genomic selection
• Yi and Xu 2008 (Genetics)
• Propose sampling σgi2 from an exponential
distribution (Bayesian LASSO)
2/22
2~)( gieP gi
λσλσ
−
-15 -10 -5 0 5 10 15
0.0
0.1
0.2
0.3
0.4
De
nsity
0 2 4 6 8 10
0.0
0.2
0.4
0.6
0.8
N = 10000 Bandwidth = 0.1188
De
nsity
Distribution of σgj2 --���� Distribution of gj
![Page 135: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/135.jpg)
Genomic selection
• Bayesian LASSO
–P(σgi2|gi)
–P(λ2|y,µ,g, σe2, σg
2)
2
2
22
222,(~,,,,| λ
σλλσµσ
j
e
ejg
InvGaussgy
)2/,(~,,,,|1
2222 ∑=
++p
j
jje bapgammagy σσσµλ
![Page 136: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/136.jpg)
Genomic selection
• Bayesian C∏ (Habier et al 2011)
• Two criticisms of BayesB
–Posterior of locus-specific variance has only one additional degree of freedom, compared to its prior regardless of the number of genotypes, so
– Degree of shrinkage of depends strongly on prior
– Little information coming from data
–∏ is treated as known, not estimated from the data
![Page 137: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/137.jpg)
Genomic selection
• Bayesian C∏ (Habier et al 2011)
• Use a common σgi2 across all SNP
–Many degrees of freedom from data
–A “BLUP” for SNP in model
• Estimate ∏ from data
–Sample from
• Beta(K - m(t) + 1, m(t) + 1).
• Where K is number of SNP, m(t) is the number of SNP in the model at iteration t (eg. Those not set to zero)
![Page 138: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/138.jpg)
Genomic selection
• Bayesian C∏ (Habier et al 2011)
– Accuracy in German Holstein Friesian data set
• Little improvement in accuracy
• But can draw inferences about trait architecture?
Trait GBLUP BayesA BayesB BayesCpi
Milk Yield 0.48 0.48 0.40 0.43
Fat Yield 0.51 0.56 0.52 0.54
Protein Yield 0.21 0.22 0.17 0.21
Somatic cells 0.17 0.17 0.12 0.14
![Page 139: Genomic Selection in the era of Genome sequencingsnp.toulouse.inra.fr/~alegarra/ben_hayes_course/... · Genomic selection with BLUP • An equivalent model • If there are many QTLs](https://reader036.vdocument.in/reader036/viewer/2022071019/5fd2a29eb190261ee73cf441/html5/thumbnails/139.jpg)
Genomic selection
• Methods for deriving prediction equation differ in assumptions about distribution of QTL effects
– BLUP = normal distribution with known variance
– Ridge regression = normal distribution with prior assumption about variance
– BayesA = t-distribution, degree of shrinkage known a-priori, or sampled
– BayesB = mixture distribution, many effects zero
– BayesianLASSO, double exponential distribution of effects
– Bayesian C∏, estimate ∏ from data, common variance across SNP