Builds the Y and X arguments of lmm_animal,
lm_phen_cov and lmm_animal_gibbs from a formula
and a data frame, so a model written the usual way (y ~ sex + age, or
cbind(x, y, z) ~ sex + age) can be fitted without building the
matrices by hand.
Usage
animal_design(
formula,
data,
na.action = stats::na.omit,
pedigree = NULL,
genotypes = NULL,
Gamma = NULL,
groups = NULL,
...
)Arguments
- formula
Model formula. The left-hand side is a single response or
cbind()of several traits; the right-hand side lists the fixed effects.- data
Data frame holding the variables in
formula. Whenpedigreeorgenotypesare used, its row names must be the individual IDs (rownames(data) <- data$id); an error says so when they were left to R.- na.action
How to handle rows with missing predictors, passed to
model.frame; the default drops them, since a row with an unknown fixed effect cannot be placed in the design. Missing values in the response are kept: a trait that was not measured on an individual is information the model can use, andlmm_animal()integrates it out.- pedigree
Optional pedigree, used to compute the additive relationship matrix
AwithmakeA. A data frame whose first three columns are the individual, its sire and its dam, with0,NAor""for unknown parents. The pedigree may contain ancestors that are not indata.- genotypes
Optional genotype matrix, used to compute a genomic relationship matrix (see
genomic_relationship): individuals in rows, with their IDs as row names, and biallelic markers in columns, coded 0, 1, 2.- Gamma
Optional relationship matrix among pedigree founders, or among founder groups when
groupsis given; passed tomakeA. Only used withpedigree.- groups
Optional founder-to-group assignment passed to
makeA; only used withpedigreeandGamma.- ...
Further arguments passed to
genomic_relationship, such asp,maforblend.
Value
A list with
- Y
response matrix, one column per trait, named after the response(s) in the formula;
- X
fixed-effects design matrix, with columns named after the terms in the formula;
- A
relationship matrix, present only when
pedigreeorgenotypesis given, restricted to the individuals inYand in the same row and column order.
Rows are in the order of data. attr(, "individuals") holds
the row names of data for the rows kept, so naming the rows with the
individual IDs lets a relationship matrix A be matched to
Y: lmm_animal puts a named A in the row order of
Y and errors if the names disagree.
Details
Random-effect terms such as (1 | animal) are ignored, so
formulas written for lme4 can be reused unchanged: relatedness is
supplied through A, which is passed to the model directly.
The result satisfies what the model functions expect: Y is numeric
and X has full column rank. Factor predictors are dummy coded and
the intercept is kept unless the formula removes it, so the columns of
X match the beta draws reported by a fit. Removing the
intercept is only allowed when the remaining columns still sum to a
constant, as the 0/1 indicators of a factor do, because the models centre
the responses and report their mean back through the design; a formula such
as y ~ 0 + x is an error.
When pedigree or genotypes is given, A is computed for
all individuals in the pedigree or genotype matrix and then subset to the
rows of Y, so ancestors without records and individuals with missing
phenotypes are handled: every row of Y must appear in the pedigree
or in the row names of the genotype matrix. With pedigree,
Gamma and groups allow related founders or metafounder
groups; see makeA.
If the response has three or more traits and its first principal component
explains at least 80% of the variance, a warning suggests fitting size and
shape scores instead (see size_shape);
pc1_variance_share reports that share. If the diagnostic
cannot be computed, a warning explains why; the design is still returned.
Examples
dat <- data.frame(trait = rnorm(6), x = rnorm(6),
sex = rep(c(0, 1), each = 3), id = 1:6)
rownames(dat) <- dat$id
design <- animal_design(trait ~ sex + x, dat)
design$Y
#> trait
#> 1 -1.400043517
#> 2 0.255317055
#> 3 -2.437263611
#> 4 -0.005571287
#> 5 0.621552721
#> 6 1.148411606
design$X
#> (Intercept) sex x
#> 1 1 0 -1.8218177
#> 2 1 0 -0.2473253
#> 3 1 0 -0.2441996
#> 4 1 1 -0.2827054
#> 5 1 1 -0.5536994
#> 6 1 1 0.6289820
#> attr(,"assign")
#> [1] 0 1 2
# formulas written for lme4 work as they are; the random term is ignored
animal_design(trait ~ sex + (1 | id), dat)
#> $Y
#> trait
#> 1 -1.400043517
#> 2 0.255317055
#> 3 -2.437263611
#> 4 -0.005571287
#> 5 0.621552721
#> 6 1.148411606
#>
#> $X
#> (Intercept) sex
#> 1 1 0
#> 2 1 0
#> 3 1 0
#> 4 1 1
#> 5 1 1
#> 6 1 1
#> attr(,"assign")
#> [1] 0 1
#>
#> attr(,"individuals")
#> [1] "1" "2" "3" "4" "5" "6"
# relatedness from a pedigree (first three columns: id, sire, dam)
ped <- data.frame(id = 1:6, sire = c(0, 0, 1, 1, 4, 4),
dam = c(0, 0, 2, 2, 3, 5))
design <- animal_design(trait ~ sex, dat, pedigree = ped)
round(design$A, 2)
#> 1 2 3 4 5 6
#> 1 1.0 0.0 0.50 0.50 0.50 0.50
#> 2 0.0 1.0 0.50 0.50 0.50 0.50
#> 3 0.5 0.5 1.00 0.50 0.75 0.62
#> 4 0.5 0.5 0.50 1.00 0.75 0.88
#> 5 0.5 0.5 0.75 0.75 1.25 1.00
#> 6 0.5 0.5 0.62 0.88 1.00 1.38
# related founders: pass their relationship matrix as Gamma
Gamma <- matrix(c(1, 0.5, 0.5, 1), 2,
dimnames = list(c("1", "2"), c("1", "2")))
design <- animal_design(trait ~ sex, dat, pedigree = ped, Gamma = Gamma)
round(design$A, 2)
#> 1 2 3 4 5 6
#> 1 1.00 0.50 0.75 0.75 0.75 0.75
#> 2 0.50 1.00 0.75 0.75 0.75 0.75
#> 3 0.75 0.75 1.25 0.75 1.00 0.88
#> 4 0.75 0.75 0.75 1.25 1.00 1.12
#> 5 0.75 0.75 1.00 1.00 1.38 1.19
#> 6 0.75 0.75 0.88 1.12 1.19 1.50
# relatedness from genotypes (individuals in rows, markers in columns)
M <- matrix(rbinom(6 * 40, 2, 0.3), 6,
dimnames = list(as.character(1:6), NULL))
design <- animal_design(trait ~ sex, dat, genotypes = M, blend = 0.01)
dim(design$A)
#> [1] 6 6
