fitMixed: fitMixed

In BGLRutils: Utilities for Bayesian Generalized Linear Regression

Description

This function fits a mixed model. See details section.

Usage

  fitMixed(y, X=NULL, Z=NULL, A=NULL,
           d=NULL,U=NULL, BLUE=TRUE, 
           BLUP=TRUE,method="ML", 
           lambda_ini=NULL, 
           tol=1e-4,maxiter=100)

Arguments

y	(numeric, n) the data-vector (NAs not allowed).
X	numeric incidence matrix for fixed effects, dimmension n x c (NAs not allowed).
Z	numeric incidence matrix for random effects, dimmension n x q (NAs not allowed).
A	numeric matrix for modelling covariance for y, for example in the animal model it can be a numeric relationship matrix derived from a pedigree. Dimmension n x n.
d	eigenvalues for G matrix. G = ZAZ'=UDU'.
U	eigenvector for G matrix. G = ZAZ'=UDU'.
BLUE	logical, if TRUE the BLUPs are computed.
BLUP	logical, if TRUE the BLUPs are computed.
method	estimation method for the variance components, can be maximum likelihood or restricted maximum likelihood. The detault is "ML" and is the only one implemented now.
lambda_ini	Initial value for varU/varE.
tol	tiny real number for declaring convergence.
maxiter	integer number for specifying the number of iterations.

Details

This function fits a mixed model

y=X*Beta + Z*u + e

where y is the vector of phenotypes, X is the matrix for FIXED EFFECTS, Z matrix connection phenotypes and genotypes, u ~ N(0, varU*A), A a genomic relationship matrix or pedigree and e~ N(0, varE*I).

The model is fitted using the algorithm described in Zhou and Stephens, 2012.

Value

A list with the following components:

varE	varE.
varU	varU.
Beta	Beta.
u	u.
message	string that informs if the method converged or not.
method	method used for computing variance components.

Author(s)

Paulino Perez, Gustavo de los Campos

References

Zhou, X. and Stephens, Matthew. 2012. Genome-wide efficient mixed-model analysis for association studies. Nature Genetics. 47(7) 821-824.

Examples

## Not run: 

library(BGLR)
library(BGLRutils)
setwd(tempdir())

#Examples
#1) Wheat dataset
data(wheat)
A=wheat.A
y=wheat.Y[,1]
ETA=list(list(K=A,model="RKHS"))
set.seed(123)
fm=BGLR(y=y,ETA=ETA,nIter=5000)

#fm_mixed=fitMixed(y,A=A)
out=eigen(A)
fm_mixed=fitMixed(y=y,A=A,d=out$values,U=out$vectors)

plot(fm$ETA[[1]]$u,fm_mixed$u)
cat("fm_mixed$varE=",fm_mixed$varE,"\n")
cat("fm$varE=",fm$varE,"\n")
cat("fm_mixed$varU=",fm_mixed$varU,"\n")
cat("fm$ETA[[1]]$varU=",fm$ETA[[1]]$varU,"\n")

#mouse dataset
data(mouse)
A=mouse.A
y=mouse.pheno$Obesity.BMI
ETA=list(list(K=A,model="RKHS"))
set.seed(123)

fm=BGLR(y=y,ETA=ETA,nIter=5000)

#fm_mixed=fitMixed(y,A=A)
out=eigen(A)
fm_mixed=fitMixed(y=y,A=A,d=out$values,U=out$vectors)

plot(fm$ETA[[1]]$u,fm_mixed$u)
cat("fm_mixed$varE=",fm_mixed$varE,"\n")
cat("fm$varE=",fm$varE,"\n")
cat("fm_mixed$varU=",fm_mixed$varU,"\n")
cat("fm$ETA[[1]]$varU=",fm$ETA[[1]]$varU,"\n")


## End(Not run)

BGLRutils documentation built on May 2, 2019, 4:48 p.m.