Quantitative genetic (animal) model example in R

Gregor Gorjanc
[email protected]

April 27, 2020

Introduction

The following is just a quick introduction to quantitative genetic model, which is usually called ani-
mal model in animal breeding scenario. This model provides inferences on parameters such as genetic
(additive/breeding, dominance, . . . ) values and possibly also co-variance components (additive genetic
variance, heritability, . . . ). Very nice introduction to this topic is in Mrode (2005), which also gives a list
of key references. We use example from this book and will therefore be very brief.

This note is mainly for educational purposes. There are quite some programs (e.g. Druet and Ducrocq
(2006) mentions ASReml , BGF90 , DFREML, DMU , MATVEC , PEST/VCE and WOMBAT )
that can fit animal models in a general manner and we suggest to take a look at them instead of trying
to reinvent the wheel in R.

In short animal model is an example of a mixed model:

y = Xb + Zu + e,

where y represents a vector of observed (measured) phenotype values, b and u are vectors of unknown
parameters for “fixed” and “random” effects, while X and Z are corresponding design matrices and finally
e is a vector of residuals. Assuming normal density for y the following standard assumptions are taken:

   
y Xb V ZG R
0 0
 u  ∼ N  0 , GZ G 0  , V = ZGZ + R
   

e 0 R 0 R

Up to now all this is as in usual mixed model. Genetic aspect comes from specification of covariance
matrix between elements of u, which usually represents sum of additive effects of genes of individuals in
the pedigree. For a univariate model the covariance matrix of additive effect can be written as G = Aσu2 ,
where A is additive/numerator relationship matrix (Wright, 1922) and σu2 is additive genetic variance
(Falconer and Mackay, 1996).

1
Mixed model equations (MME)

Solution for b i.e. (E)BLUE and u i.e. (E)BLUP can be obtained from (Henderson, 1949; Goldberger,
1962; Henderson, 1963):

−
b̂ = XV−1 X XV−1 y,

and

0
 
û = GZ V−1 y − Xb̂ ,

but in a case with a lot of records the size of V is huge and its direct inverse prohibitive if possible at all.
Henderson (1950) presented the solution to this problem with so called mixed model equations:

0 0
! ! 0
!
X R−1 X X R−1 Z b̂ X R−1 y
0 0 = 0 .
Z R−1 X Z R−1 Z + G−1 û Z R−1 y

Data

We will use pedigree and data example from Mrode (2005). Example shows a beef breeding scenario with
8 individuals (animals), where 5 of them have phenotype records (pre-weaning gain in kg) and 3 three of
them are without records and link others through the pedigree.

> library(GeneticsPed)
> data(Mrode3.1)
> (x <- Pedigree(x=Mrode3.1, subject="calf", ascendant=c("sire", "dam"),
+ ascendantSex=c("Male", "Female"), sex="sex"))

calf sex sire dam pwg

1 S4 Male S1 <NA> 4.5
2 S5 Female S3 S2 2.9
3 S6 Female S1 S2 3.9
4 S7 Male S4 S5 3.5
5 S8 Male S3 S6 5.0

The model

For this baby BLUP example we will postulate the following model:

yij = si + aj + eij ,

where yij is pre-weaning gain (kg) of calf j of sex j; si are parameters of sex effect, while aj are parameters
of additive genetic effect for pre-weaning gain and finally eij is residual. Variances for aj and eij are
assumed as G = Aσa2 with σa2 = 20 kg 2 and R = Iσe2 with σe2 = 40 kg 2 .
2
Setting up the MME

Observed/measured phenotype records:

> (y <- x$pwg)

[1] 4.5 2.9 3.9 3.5 5.0

Design matrix (X) for sex effect:

> X <- model.matrix(~ x$sex - 1)

> t(X)

1 2 3 4 5
x$sexFemale 0 1 1 0 0
x$sexMale 1 0 0 1 1
attr(,"assign")
[1] 1 1
attr(,"contrasts")
attr(,"contrasts")$`x$sex`
[1] "contr.treatment"

Design matrix (Z) for additive genetic effect. Note that first three columns do not have indicators since
these columns are for individuals without phenotype records and apear in the model only through the
pedigree.

> (Z <- model.matrix(object=x, y=x$pwg, id=x$calf))

S2 S1 S3 S4 S5 S6 S7 S8
1 0 0 0 1 0 0 0 0
2 0 0 0 0 1 0 0 0
3 0 0 0 0 0 1 0 0
4 0 0 0 0 0 0 1 0
5 0 0 0 0 0 0 0 1

Left hand side (LHS) of MME without G−1 :

> LHS <- rbind(cbind(t(X) %*% X, t(X) %*% Z),

+ cbind(t(Z) %*% X, t(Z) %*% Z))
> ## or more efficiently
> (LHS <- rbind(cbind(crossprod(X), crossprod(X, Z)),
+ cbind(crossprod(Z, X), crossprod(Z))))

3
 x$sexFemale x$sexMale S2 S1 S3 S4 S5 S6 S7 S8
x$sexFemale 2 0 0 0 0 0 1 1 0 0
x$sexMale 0 3 0 0 0 1 0 0 1 1
S2 0 0 0 0 0 0 0 0 0 0
S1 0 0 0 0 0 0 0 0 0 0
S3 0 0 0 0 0 0 0 0 0 0
S4 0 1 0 0 0 1 0 0 0 0
S5 1 0 0 0 0 0 1 0 0 0
S6 1 0 0 0 0 0 0 1 0 0
S7 0 1 0 0 0 0 0 0 1 0
S8 0 1 0 0 0 0 0 0 0 1

σe2
and adding G−1 , which is in this case A−1 α and α = σa2 = 40
20 = 2.

> ## We want Ainv for all individuals in the pedigree not only individuals
> ## with records
> x <- extend(x)
> Ainv <- inverseAdditive(x=x)
> sigma2a <- 20
> sigma2e <- 40
> alpha <- sigma2e / sigma2a
> q <- nIndividual(x)
> p <- nrow(LHS) - q
> (LHS[(p + 1):(p + q), (p + 1):(p + q)] <-
+ LHS[(p + 1):(p + q), (p + 1):(p + q)] + Ainv * alpha)

S2 S1 S3 S4 S5 S6 S7 S8
S2 4 1.000000 1 0.000000 -2 -2 0 0
S1 1 3.666667 0 -1.333333 0 -2 0 0
S3 1 0.000000 4 0.000000 -2 1 0 -2
S4 0 -1.333333 0 4.666667 1 0 -2 0
S5 -2 0.000000 -2 1.000000 6 0 -2 0
S6 -2 -2.000000 1 0.000000 0 6 0 -2
S7 0 0.000000 0 -2.000000 -2 0 5 0
S8 0 0.000000 -2 0.000000 0 -2 0 5

Right hand side (RHS) of MME:

> RHS <- rbind(t(X) %*% y,

+ t(Z) %*% y)
> ## or more efficiently
> RHS <- rbind(crossprod(X, y),
+ crossprod(Z, y))
> t(RHS)

x$sexFemale x$sexMale S2 S1 S3 S4 S5 S6 S7 S8
[1,] 6.8 13 0 0 0 4.5 2.9 3.9 3.5 5
4
Solution

> sol <- solve(LHS) %*% RHS

> ## or more efficiently
> sol <- solve(LHS, RHS)
> t(sol)

x$sexFemale x$sexMale S2 S1 S3 S4 S5
[1,] 3.40443 4.358502 -0.0187701 0.09844458 -0.0410842 -0.008663123 -0.1857321
S6 S7 S8
[1,] 0.1768721 -0.2494586 0.1826147

That’s all folks! Well, all for the introduction. There are numerous issues covered in the literature. A
good starting point is Mrode (2005) as already mentioned in the beginning.

R Session information

> toLatex(sessionInfo())

ˆ R version 4.0.0 (2020-04-24), x86_64-pc-linux-gnu

ˆ Locale: LC_CTYPE=en_US.UTF-8, LC_NUMERIC=C, LC_TIME=en_US.UTF-8, LC_COLLATE=C,

LC_MONETARY=en_US.UTF-8, LC_MESSAGES=en_US.UTF-8, LC_PAPER=en_US.UTF-8, LC_NAME=C,
LC_ADDRESS=C, LC_TELEPHONE=C, LC_MEASUREMENT=en_US.UTF-8, LC_IDENTIFICATION=C

ˆ Running under: Ubuntu 18.04.4 LTS

ˆ Matrix products: default

ˆ BLAS: /home/biocbuild/bbs-3.11-bioc/R/lib/libRblas.so

ˆ LAPACK: /home/biocbuild/bbs-3.11-bioc/R/lib/libRlapack.so

ˆ Base packages: base, datasets, grDevices, graphics, methods, stats, utils

ˆ Other packages: GeneticsPed 1.50.0, MASS 7.3-51.6

ˆ Loaded via a namespace (and not attached): combinat 0.0-8, compiler 4.0.0, gdata 2.18.0,
genetics 1.3.8.1.2, gtools 3.8.2, mvtnorm 1.1-0, tools 4.0.0

References
Druet, T. and Ducrocq, V. (2006). Innovations in software packages in quantitative genetics. In 8th
World Congress on Genetics Applied to Livestock Production, Belo Horizonte, 2006-08-13/18. Brazilian
Society of Animal Breeding. Communication 27-10.

Falconer, D. S. and Mackay, T. F. C. (1996). Introduction to Quantitative Genetics. Longman, Essex,

U.K., 4th ed. edition.
5
Goldberger, A. S. (1962). Best linear unbiased prediction in the generalized linear regression model. J.
Am. Stat. Assoc., 57(6):369–375.

Henderson, C. R. (1949). Estimation of changes in herd environment. 32:709. (Abstract).

Henderson, C. R. (1950). Estimation of genetic parameters. Ann. Math. Stat., 9:309.

Henderson, C. R. (1963). Selection index and expected genetic advance. In Hanson, W. D. and Robinson,
H. F., editors, Statistical genetics and plant breeding, number 982, pages 141–163. National academy
of sciences and national research council, Washington DC.

Mrode, R. A. (2005). Linear models for the prediction of animal breeding values. CAB International,
Wallingford, Oxon OX10 8DE, UK, 2 edition.

Wright, S. (1922). Coefficients of inbreeding and relationship. Am. Nat., 56:330–338.