Genomic Correlation: Harnessing the benefits of combining two unrelated populations for genomic selection - Toni Reverter

Genomic Correlation: Harnessing the benefit of combining two unrelated populations for genomic selection

CSIRO AGRICULTURE

Toni Reverter – OCE Symposium, Genome to Phenome, 25-27 March 2015.

LR Porto-Neto, W Barendse, JM Henshall, SM McWilliam,SA Lehnert and A Reverter Currently under review in

Genetics Selection Evolution

Toni Reverter – OCE Symposium: Genome to Phenome, March 2015, UQ Brisbane.

Introduction

• Genomic selection uses a reference population to build & calibrate the genomic-based predictions of genetic value (GEBV).

• The larger the reference population, the more accurate the predictions.

• Also impacting accuracy:

• Heritability• LD decay• Relationship with target population• Selection of markers


Introduction


Introduction


Introduction

• One way to increase the size of the reference population is to merge data from different populations (breeds) Mixed Results


Introduction

• Assumption #1: A genetic correlation (>0!) exists for the phenotype of interest in the two populations

The stronger this correlation the higher the benefit.

• Assumption #2: The genomic relationship matrix (GRM) allows for estimation of this “genomic” correlation.

• Assumption #3: This genomic correlation can be manipulated by selecting (via GWAS) which SNP to use when building the GRM.


Objectives

• Estimate the genomic correlation for 5 phenotypes related to tropical adaptation in beef cattle

• To show that a careful selection of the SNP based on LD phase (same or different) and having highly significant effect impacts on the estimates of genomic correlation and accuracy of GEBV

• Data: 1,829 Brahman (BB) and 1,973 Tropical Composite (TC) cows and bulls genotyped for 71,726 SNP highly polymorphic in Bos indicus cattle.


Methods

Tropical Composite (TC) Brahman (BB)


Flowchart

REFERENCE POPULATION817 BB + 1,028 TC Cows

5 Phenotypes71,726 SNPs

GWAS AnalysesNRM + SNP

1 2




Whole GRM: GW


Same Effect GRM: GS Different Effect GRM: GD

1 2

Flowchart




Whole GRM: GW

Bi-Variate AnalysesNRM + GRM



GenomicCorrelations

Missing Heritability

1 2

3

Flowchart




Whole GRM: GW

Bi-Variate AnalysesNRM + GRM



VALIDATION POPULATION1,012 BB + 945 TC Bulls

5 Phenotypes

GenomicCorrelations

Missing Heritability

Uni-Variate GBLUP AnalysesGRM only

GEBVAccuracies

1 2

3 4

5

Flowchart


Relationship Matrices

Brahman

Tropical Composite

NRM

GRM


Models

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

e

e

u

u

M0

0M

a

a

Z0

0Z

β

β

X0

0X

y

y

REML Estimation of genetic parameters and fraction of missing heritability

I00000

0I0000

00GG00

00GG00

0000A0

00000A

e

e

u

u

a

a

2e

2e

2uu

u2u

2a

2a

TC

BB

TC

BB

TC

BB

TC

BB

TCTCBB,

TCBB,BB

TC

BB

V

Fixed Effects Pedigree (NRM) Genotype (GRM)


Models

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

e

e

u

u

M0

0M

a

a

Z0

0Z

β

β

X0

0X

y

y


I00000

0I0000

00GG00

00GG00

0000A0

00000A

e

e

u

u

a

a

2e

2e

2uu

u2u

2a

2a

TC

BB

TC

BB

TC

BB

TC

BB

TCTCBB,

TCBB,BB

TC

BB

V2u

2u

u

G

TCBB

TCBB,

r

Genomic Correlation



Models

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

TC

BB

e

e

u

u

M0

0M

a

a

Z0

0Z

β

β

X0

0X

y

y


I00000

0I0000

00GG00

00GG00

0000A0

00000A

e

e

u

u

a

a

2e

2e

2uu

u2u

2a

2a

TC

BB

TC

BB

TC

BB

TC

BB

TCTCBB,

TCBB,BB

TC

BB

V2u

2u

u

G

TCBB

TCBB,

r

Genomic Correlation

2a

2u

2uBB

BBBB

BB1

missC

Missing h2



GRMs

1. GW = the GRM built with the Whole set of 71,726 SNP genotypes;

2. GS = the GRM built with 16,207 SNP having effects in the Same direction in both breeds and across all five phenotypes;

3. GD = the GRM built with 16,951 SNP having effects in Different direction in both breeds and across all five phenotypes.

1 2 3 4 5

16,207 Same 75.18 19.70 4.15 0.89 0.08

16,951 Diff. 76.91 19.26 3.41 0.38 0.05

% SNPs affecting 1 to 5 phenotypes

For each phenotype select the 10% (or 7,200) most significant SNP and in the same direction in both breeds. Merge the 5 lists into a single list, namely “list-of-same”.

Similarly, create the “list-of-different”. Finally, remove any SNP in the overlap of “list-of-same” and “list-of-different”.


Results

GW

GD

GS


Results: ACC of GEBV


Results: ACC of GEBV


Conclusions

1. For the merging of two populations into a single reference to be beneficial, there must be a correlation between the same phenotype in the two populations.

2. There is benefit in including relevant markers and excluding irrelevant ones.

3. This careful selection of markers impacts on: • The estimate of the genomic correlation.• The accuracy of GEBV.

4. Future work:• Mechanism for “careful” selection• Allele frequencies used in GRM• What if highly unbalanced population sizes?

CSIRO AGRICULTURE

Thank you!

Genomic Correlation: Harnessing the benefits of combining two unrelated populations for genomic selection - Toni Reverter

Science

uq brisbane

gs different effect

introduction genomic

flowchart reference

snps gwas analyses nrm

gw gwas analyses nrm

gd genomic correlations

genetic correlation