Cranial Euclidean distances and individual information for opossums from the
colony of the South Texas Diabetes and Obesity Institute, analysed by
Porto et al. (2016). The traits are the same 35 landmarks as
ratones_morpho, so the two datasets can be compared directly.
Format
Two objects.
opossuma data frame with 553 rows and 38 columns:
id,sex,agein days, and the 35 cranial distances as measured, named after the landmarks they span, with underscores as inratones_morpho. Row names are the individual IDs. 22 individuals are missing every trait.lmm_animalaccepts fully missing response rows, but they must be removed before fittinglm_phen_cov. 505 individuals have no missing value in any trait or covariate.opossum_grma 553 by 553 genomic relationship matrix computed from 3,696 markers as \(ZZ^\top / (2\sum_j p_j(1-p_j))\) (VanRaden 2008), \(Z\) being the allele-count matrix centred by twice the sample allele frequencies \(p_j\). It equals
genomic_relationship()applied to the markers, and it is on the scale of a pedigree relationship matrix, soGfitted on it is the additive genetic covariance of the population. Rows and columns are named with the individual IDs ofopossum, so the rows of the subset being analysed can be passed as theAargument oflmm_animaldirectly, as in the example. The full matrix is only positive semidefinite, so passblend_A = TRUEif it is given whole. The diagonal runs from 0.64 to 1.70 with a mean of 1.05, and the off-diagonal entries average zero, since the allele frequencies come from the same animals. Versions up to 0.8.1 shipped the same matrix divided by the number of markers instead, 2.86 times smaller, which inflatedGand the heritabilities fitted on it.
Source
Porto, A., Schmelter, R., VandeBerg, J. L., Marroig, G., & Cheverud, J. M. (2016). Evolution of the genotype-to-phenotype map and the cost of pleiotropy in mammals. Genetics, 204(4), 1601-1612. doi:10.1534/genetics.116.189431 . Data deposited on figshare as “Data - Genetics - Porto et al 2016”, doi:10.6084/m9.figshare.4055961 .
The coded marker files in that archive carry no individual identifiers, so
the animals are matched to the cranial measurements by row order. The two
sources agree in length at 553, and the identities are attached here as the
row and column names of opossum_grm.
Details
Unlike the mice, these animals carry marker genotypes, so the relationship matrix is genomic rather than genealogical. That matters for these data: the colony pedigree shipped with the original archive gives every pair a kinship coefficient between 0.69 and 0.93, which leaves additive variance indistinguishable from a shared colony effect. The genomic matrix separates relatives from non-relatives and makes the genetic covariance identifiable.
The distances are stored as measured. Porto et al. (2016) analysed them untransformed, and the phenomenon is confirmed by their published phenotypic covariance, whose diagonal correlates 0.834 with the variances of the raw distances against 0.046 with those of their logarithms.
References
Porto, A., Schmelter, R., VandeBerg, J. L., Marroig, G., & Cheverud, J. M. (2016). Evolution of the genotype-to-phenotype map and the cost of pleiotropy in mammals. Genetics, 204(4), 1601-1612. doi:10.1534/genetics.116.189431
Examples
data(opossum)
dim(opossum)
#> [1] 553 38
dim(opossum_grm)
#> [1] 553 553
traits <- setdiff(names(opossum), c("id", "sex", "age"))
if (FALSE) { # \dontrun{
# Phenotypic covariance of four traits, no pedigree needed
lm_phen_cov(as.matrix(opossum[, traits[1:4]]),
model.matrix(~ sex + age, opossum))
# Genomic animal model on the same four traits
design <- animal_design(
stats::reformulate(c("sex", "age"),
response = sprintf("cbind(%s)", paste(traits[1:4],
collapse = ", "))),
data = opossum, pedigree = NULL)
lmm_animal(design$Y, design$X, opossum_grm[rownames(design$Y),
rownames(design$Y)])
} # }
