Realized Genomic Relationships and Genomic BLUP Fikret Isik Associate Professor September 16,...
-
Upload
maximillian-kelly -
Category
Documents
-
view
223 -
download
1
Transcript of Realized Genomic Relationships and Genomic BLUP Fikret Isik Associate Professor September 16,...
Genomic BLUP - UofCopenhagen 1
Realized Genomic Relationships and Genomic BLUP
Fikret IsikAssociate Professor
September 16, 2013
Cooperative Tree Improvement ProgramNorth Carolina State University, Raleigh, USA
Genomic BLUP - UofCopenhagen 2
Outline
• Background• Realized genomic relationships
– G matrix– H matrix
• Genomic BLUP• Empirical Examples from Pinus taeda L.
September 16, 2013
Genomic BLUP - UofCopenhagen 3
Expected covariances
• Additive genetic relationships (covariances) derived from a pedigree are based on probabilities that gene pairs are identical by descent (IBD)
• For example, the average genetic covariance between full-sibs is 0.5 because full-sibs are expected to share 50% of their genome that is IBD
September 16, 2013
Genomic BLUP - UofCopenhagen 4
Genetic merit
• In the classic “infinitesimal model” of quantitative genetics, breeding value is considered to be the sum of thousands of allelic effects.
September 16, 2013
Genomic BLUP - UofCopenhagen 5
Traditional genetic evaluation
• Infinitesimal model has been very successful to predict genetic merit of individuals in animal- and plant-improvement programs
• This model does not trace individual alleles (black box) (VanRaden 2008).
September 16, 2013
Genomic BLUP - UofCopenhagen 6
Tracing loci
• “Remember that all models are wrong; the practical question is how wrong do they have to be to not be useful” George Box.
• In real genomes, those alleles are physically located at loci whose transmission can be traced through genetic markers.
September 16, 2013
Genomic BLUP - UofCopenhagen 7
The real deficiency of model
• Genetic relationships derived from pedigree ignore the random sampling of the two possible alleles from each parent at each locus during meiosis (Avendano et al. 2005).
• In the absence of phenotype, selection is not possible in a cross.
September 16, 2013
Genomic BLUP - UofCopenhagen 8
Mendelian segregation effect (m)
• When gametes are produced (by meiosis) allele pairs segregate, leaving each cell with a single allele (Mendel’s law of segregation).
• Each progeny receives 50% of parental DNA, random sampling of parent alleles at each locus during meiosis
• The genetic merit: 0.5 (uj + uk) + mi where j, k are parents of i
September 16, 2013
Genomic BLUP - UofCopenhagen 9
Genetic similarities
• van Arendonk et al. (1994) suggested that large numbers of DNA markers covering the genome could measure genetic similarity more accurately than a pedigree-based relationship
• because the genetic covariances would be based on the actual proportion of the genome that is IBD between any two individuals.
September 16, 2013
Genomic BLUP - UofCopenhagen 10
Realized genomic relationships
• Genetic markers could estimate proportion of chromosome segments shared by individuals including identification of genes IBS (VanRaden, 2008)
September 16, 2013
Genomic BLUP - UofCopenhagen 11
Genomic predictions
• Selection based on realized genomic relationships can produce more accurate predictions than the pedigree-based method
• because genomic selection can exploit variation created by Mendelian segregation during gamete formation (Goddard and Hayes 2007)
September 16, 2013
Genomic BLUP - UofCopenhagen 12
Genomic predictions (cont.)
• Such methods do not require known location of markers in the genome or
• do not require estimation of relative effects of individual QTL on the trait.
September 16, 2013
Genomic BLUP - UofCopenhagen 13
Matrix of gene content (M) The product of M matrix with its transpose M´ is MM’ matrix
• Diagonal: Counts the # of homozygous loci for each individual.
• Off-diagonal: Measure the number of alleles shared by relatives
individual 1 individual 2 individual 3
(VanRaden, 2008, Forni et al. 2011)
September 16, 2013
Genomic BLUP - UofCopenhagen 14
Realized genomic relations matrix
(VanRaden 2008)
• Dividing by scales G to be analogues to the A matrix
• p_i are the observed MAF of all genotyped individuals regardless of inbreeding and selection
September 16, 2013
ZZ’ = (M – P) (M – P)’
Genomic BLUP - UofCopenhagen 15
Assumptions of GS
• QTL explaining genetic variation are in LD with genetic markers (Meuwissen et al. 2001).
• We do not know the frequency of alleles IBS are actually are IBD, especially in outbred populations (Legarra et al. 2009).
September 16, 2013
Genomic BLUP - UofCopenhagen 16
Inverse of G matrix
• The genomic matrix is positive semidefinite but it can be singular (no unique solution) if– Number of loci is limited – Subjects have identical genotypes across all loci– Number of markers is smaller than the number of
individuals genotyped
September 16, 2013
Genomic BLUP - UofCopenhagen 17
Inverse of G matrix
• To avoid potential problems G can be weighted
• Gr is unweighted genomic matrix• A is numerator relationship matrix among only
genotyped animals• w is weight - This value is not critical between
values of 0.95 and 0.98 (Aguilar et al. 2010)
September 16, 2013
Genomic BLUP - UofCopenhagen 18
Hybrid genetic relationship matrices
• Genotyping may not be reasonable for all the population due to high cost and logistic limitations, particularly for tree breeding populations.
September 16, 2013
Genomic BLUP - UofCopenhagen 19
Hybrid genetic relationships
• vanRaden and (2008) and Legarra et al. (2009) proposed combining numerator relationships matrix (A) derived from pedigree with the genomic relationship matrix (G) into a single matrix (H = A+G ) to use in predictions.
September 16, 2013
Genomic BLUP - UofCopenhagen 20
A modified animal model
• In H matrix, genomic information is transmitted to the covariances among all non-genotyped individuals (Legarra et al. 2009).
• The H matrix is a joint distribution of genotyped and non-genotyped genetic values
September 16, 2013
Genomic BLUP - UofCopenhagen 21
Construction of H matrix
• Instead of A-1, genomic analysis uses
• is contribution of genomic relationships in H
September 16, 2013
A22 for the genotyped individuals
Misztal et al. 2009, VanRaden 2008
Genomic BLUP - UofCopenhagen 22
Predictions of genetic merit of trees using G and H matrices
September 16, 2013
Genomic BLUP - UofCopenhagen 23
Genomic BLUP
September 16, 2013
Genomic estimated breeding values using selection index equations
Markers effects can be estimated by substituting the Z’ to the leftmost G
Genomic BLUP - UofCopenhagen 24
Mixed models for GBLUP
y = Xb + Zu + e • Xb is the mean (other fixed effects could be added)• Z is incidence matrix for marker effects• u is vector of additive genetics effects that
correspond to allele substitution effects for each marker
• We let the sum Zu across all marker loci (m) to be equal to the vector of breeding values Za = u
September 16, 2013
(VanRaden 2008)
Genomic BLUP - UofCopenhagen 25
Mixed model equations for GBLUP
September 16, 2013
Lambda is defined as the sum across loci 2Σpi1-pi times the ratio of error and additive genetic variance
Genomic BLUP - UofCopenhagen 26
Emprical results from Pinus taeda L.
September 16, 2013
Genomic BLUP - UofCopenhagen 27September 16, 2013
Genomic BLUP - UofCopenhagen 28
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Additive genetic covariance
A
Pedigree-based relationships
mean = 0.54
min = 0.41
max = 0.62
N = 3,998
September 16, 2013
for 305 progeny, from 9 families
Genomic BLUP - UofCopenhagen 29
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Realized genomic covariance
G
Genomic relationships
mean = 0.53
min = 0
max = 0.95
N = 1,967
September 16, 2013
for 165 trees, from 9 families
Genomic BLUP - UofCopenhagen 30
Predictions from ABLUP-GBLUP
September 16, 2013
Zapata-Valenzuela et al. 2013
Genomic BLUP - UofCopenhagen 31
Accuracies of the predictions
Training / validation r(ABLUP) r(GBLUP)
84 / 81 0.60 0.71148 / 17 0.61 0.76
Accuracies of predictions from markers (GBLUP) are higher than accuracies of predictions from pedigree based models (ABLUP)
September 16, 2013
Zapata-Valenzuela et al. 2013 Genes Genomes Genetics.
32September 16, 2013 Genomic BLUP - UofCopenhagen
Zapata-Valenzuela et al. 2013 Genes Genomes Genetics.
Markers are capturing the Mendelian sampling effect
Genomic BLUP - UofCopenhagen 33
Predictions from blended genomic relationship in Pinus taeda
Manuscript in preparation
September 16, 2013
Funda Ogut, (NCSU Crop science)
Genomic BLUP - UofCopenhagen 34
Actual data (Mid-parent EBVs)
ABLUP: Predictions of full-sib progeny within nine families.
No phenotype was available and predictions are mid-parent breeding values
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
-3-2
-10
12
ABLUP_ide
TB
V_
ide
September 16, 2013
Genomic BLUP - UofCopenhagen 35
Actual data (Mid-parent EBVs)ABLUP: Predictions of full-sib progeny within nine families
Not much segregation within a cross
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
-3-2
-10
12
ABLUP_ide
TB
V_
ide
September 16, 2013
Genomic BLUP - UofCopenhagen 36
Actual data-TBVs and EBVsHBLUPDark blue dots: Non-genotyped
Red dots: predictions from HBLUP
-3 -2 -1 0 1 2 3
-3-2
-10
12
HBLUP_GOF
TB
V_
ide
r=0.73
September 16, 2013
HBLUP models captures Mendelian segregation effect (different BV) within full-sib crosses.
Genomic BLUP - UofCopenhagen 37
Summary
• Realized genomic relationships allow capturing the Mendelian sampling effect for within-family (forward) selection without phenotype
• An important advantage to control inbreeding and increase genetic gain across multiple generations in forest tree breeding compared to the traditional evaluation
September 16, 2013
Genomic BLUP - UofCopenhagen 38
Summary (cont.)
• HBLUP uses all the available genomic, pedigree and phenotype information in one step for genomic predictions
• Implementation is straightforward• Standard software available for linear mixed
models can be used to solve for mixed model equations while accounting for experimental design factors, such as location and age
September 16, 2013