Fit mixed-effects models incorporating pedigrees
This example illustrates how to use spaMM for quantitative genetic analyses. spaMM appears competitive in terms of speed for GLMMs with large data sets, particularly when using the PQL method, which may be a quite good approximation in such cases. For large pedigrees it may be useful to compute the inverse of the relationship matrix using some efficient ad hoc algorithm, then to provide it as argument of the fit using the covStruct(list(precision=...)) syntax. If the precision matrix is not specified, spaMM will generally evaluate it to assess whether it should use sparse-precision methods. see sparse_precision for further control of this computation, on another example from quantitative genetics.
## Not run:
if(requireNamespace("pedigreemm", quietly=TRUE)) {
## derived from help("pedigreemm")
p1 <- new("pedigree",
sire = as.integer(c(NA,NA,1, 1,4,5)),
dam = as.integer(c(NA,NA,2,NA,3,2)),
label = as.character(1:6))
A <- pedigreemm::getA(p1) ## relationship matrix
## data simulation
cholA <- chol(A)
varU <- 0.4; varE <- 0.6; rep <- 20
n <- rep*6
set.seed(108)
bStar <- rnorm(6, sd=sqrt(varU))
b <- crossprod(as.matrix(cholA),bStar)
ID <- rep(1:6, each=rep)
e0 <- rnorm(n, sd=sqrt(varE))
y <- b[ID]+e0
obs <- data.frame(y=y,IDgen=ID,IDenv=ID) ## two copies of ID for readability of GLMM results
## fits
fitme(y ~ 1+ corrMatrix(1|IDgen) , corrMatrix=A,data=obs,method="REML")
obs$y01 <- ifelse(y<1.3,0,1)
fitme(y01 ~ 1+ corrMatrix(1|IDgen)+(1|IDenv), corrMatrix=A,data=obs,
family=binomial(), method="REML")
prec_mat <- solve(A)
colnames(prec_mat) <- rownames(prec_mat) <- rownames(A) # important
fitme(y01 ~ 1+ corrMatrix(1|IDgen)+(1|IDenv) , covStruct=list(precision=prec_mat),
data=obs, family=binomial(), method="REML")
}
## End(Not run)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.