G.predict {ASRgenomics}

Generates the conditional predictions of random effects (BLUPs)

Description

Predicts random effects values for individuals with unobserved responses (here called x, a vector of length nx) based on known random effect values for individuals with observed responses (here called y, a vector of length ny). This is done using the common genomic relationship matrix \boldsymbol{G} for all individuals (full matrix of dimension (nx + ny) \times (nx + ny)).

The prediction of unobserved responses will be performed through the multivariante Normal conditional distribution. These predictions are identical to what would be obtained if the entire set of individuals (nx + ny) were included into a GBLUP animal model fit with individuals in the set x coded as missing.

The user needs to provide the matrix \boldsymbol{G} in full form. Individual names (nx + ny) should be assigned to rownames and colnames, and these can be in any order. If the variance-covariance matrix of the set y is provided, standard errors of random effects in set x are calculated.

Usage

G.predict(G = NULL, gy = NULL, vcov.gy = NULL)

Arguments

G

Input of the genomic relationship matrix \boldsymbol{G} in full form (of dimension (nx + ny) \times (nx + ny)) (default = NULL).

gy

Input of random effects (e.g. breeding values) for individuals with known values. Individual names should be assigned to rownames of this vector and be found on the matrix \boldsymbol{G} (default = NULL).

vcov.gy

The variance-covariance matrix associated with the random effects from the individuals with known values (set y, of dimension ny \times ny) (default = NULL).

Value

A data frame with the predicted random effect values for individuals with unobserved responses in the set x. If the variance-covariance matrix is provided, standard errors are included in an additional column.

Examples

## Not run: 
library(asreml) # Load asreml.

# Example: Apple data creating 100 missing observations.

# Prepare G (nx + ny).
G <- G.matrix(M = geno.apple, method = "VanRaden", sparseform = FALSE)$G
dim(G)

# Prepare subset of data.
# Select only 147 from 247 individuals from pheno.apple and geno.apple.
Gy <- G[1:147, 1:147]
phenoy <- pheno.apple[1:147, ]

# Obtain the BLUPs for the 147 individuals using ASReml-R.

# Blend Gy.
Gyb <- G.tuneup(G = Gy, blend = TRUE, pblend = 0.02)$Gb

# Get the Gy inverse
Gyinv  <- G.inverse(G = Gyb, sparseform = TRUE)$Ginv.sparse

# Fit a GBLUP model
phenoy$INDIV <- as.factor(phenoy$INDIV)
modelGBLUP <-
 asreml(
  fixed = JUI_MOT ~ 1,
  random = ~vm(INDIV, Gyinv),
  workspace = 128e06,
  data = phenoy)

# Obtain Predictions - BLUP (set y).
BLUP <- summary(modelGBLUP,coef = TRUE)$coef.random
head(BLUP)
gy <- as.matrix(BLUP[, 1])
rownames(gy) <- phenoy$INDIV

# Ready to make conditional predictions.
g.cond <- G.predict(G = G, gy = gy)
head(g.cond)

## End(Not run)