emma.ML.LRT {emma}R Documentation

Linear mixed model association via Likelihood Ratio test with ML estimates.

Description

Performs an efficient linear mixed model association mapping via Likelihood Ratio test (LR) after estimating variance component using REML.

Usage

  emma.REML.LRT (ys, xs, K, Z=diag(ncol(xs)),
  X0 = matrix(1,nrow(xs),1), ngrid=100, llim=-5, ulim=5,
  esp=1e-10, ponly = FALSE)

Arguments

ys A g by n matrix, where g is the number of response variables (or phenotypes), and n is the number of individuals
xs A m by t matrix, where m is number of indicator variables (or snps), and n is the number of strains
K A t by t matrix of kinship coefficients, representing the pairwise genetic relatedness between strains
Z A n by t incidence matrix mapping each individual to a strain. If this is NULL, n and t should be equal and an identity matrix replace Z
X0 A n by p matrix of fixed effects variables, where p is the number of fixed effects including mean and other confounding variables
ngrids Number of grids to search optimal variance component
llim Lower bound of log ratio of two variance components
ulim Upper bound of log ratio of two variance components
esp Tolerance of numerical precision error
ponly Returns p-value matrix only if TRUE
eig.L Eigenvector from Z and K used in ML estimate. If specified, it may avoid redundant computation inside the function
eig.R0 Eigenvector from X0, Z and K used in REML estimate. If specified, it may avoid redundant computation inside the function
eig.R1 Eigenvector from X1, Z and K used in REML estimate. If specified, it may avoid redundant computation inside the function Valid only when m=1

Details

The following criteria must hold; otherwise an error occurs - [# cols in ys] == [# rows in Z] == [# rows in X0] - [# cols in xs] == [# cols in Z] == [# rows in K] == [# cols in K] - rowSums(Z) should be a vector of ones - colSums(Z) should not contain zero elements - K must be a positive semidefinite matrix

Value

A list containing:

ps The g by m matrix of p-values between every pair of indicator-response variables
stats The g by m matrix of chi-square statistic values
ML0s The g by m matrix of log maximum likelihood under null hypothesis
ML1s The g by m matrix of log maximum likliehood under alternative hypothesis
vgs The g by m matrix of genetic variance components under null hypothesis
ves The g by m matrix of random variance components under null hypothesis


if ponly is TRUE, only ps is return as matrix form.

Author(s)

Hyun Min Kang h3kang@cs.ucsd.edu

References

Kang HM, Zaitlen NA, Wade CM, Kirby A, Heckerman D, Daly MJ, and Eskin E, Efficient Control of Population Structure in Model Organism Association Mapping, Genetics 178:1709-1723, 2008

See Also

emma.REML.t,emma.kinship,emma.MLE

Examples

  ## Not run: 
    ## Load data
    data(emmadat)

    ## Run EMMA
    rs <- emma.ML.LRT(emmadat$ys,emmadat$xs,emmadat$K)

    ## return p-values
    rs$ps
  ## End(Not run)

[Package emma version 1.1.2 Index]