sample.dep: Simulate a sample of groups of dependent normal observations

In warnes/exp.ssize: Estimate Effectivenes of Microarry Sample Size Estimation via Simulation

Description

Given a structure of covariance matrices to sample from, this function repeats the steps of first randomly selects covariance matrices and then simulate normal observation based on the current selected covariance until the sample of specified size is reached

Usage

sample.dep(ngenes.dep, n, var.ratio, cov.matrix, ngenes.matrix, delta, distrn = "normal")

Arguments

ngenes.dep	integer number of genes to sample the dependent expression
n	positive integer number of replicates for each gene each group (control/treatment)
var.ratio	a positive number for the ratio of variance between treatment and control
cov.matrix	a list of covariance matrices to sample the covariance from
ngenes.matrix	the number of covariance matrices in the list of cov.matrix
delta	effect size for sample size calculation the absolute distance from zero for expression values of genes coming from the true alternative hypothesis group
distrn	distribution of expression data, set to be normal for here

Value

returns a matrix with ngenes.ind rows and 2*n columns with the first n columns being samples from control group and last n columns being from treatment group for each gene (row). each group of genes consists of subgroups of genes that are dependent

Author(s)

Peng Liu peng\[e-mail]

Examples

 library(MASS)
# set.seed(100)

# initial values to put in argument
n.dep <- 10; n <- 1000; delta <- 50; var.ratio <- 1;

ngenes <- 7; cov.matrix <- list(ngenes);

for ( i in c(1:ngenes)) {
    cov.matrix[[i]] <- diag(rep(i*0.1, i+1))
    # cov.matrix[[i]] <- diag(rep(i*0.1,4))
}

cov.matrix[[4]][1,1] <- cov.matrix[[5]][1,1] <- cov.matrix[[6]][1,1] <- 1
cov.matrix[[4]][2,1] <- cov.matrix[[4]][1,2] <- 0.15 # corr = 0.15 / (1*0.2) = 0.75
cov.matrix[[5]][2,1] <- cov.matrix[[5]][1,2] <- 0.5 # corr = 0.15 /(1*sqrt(0.5)) = 0.59
cov.matrix[[7]][1,1] <- 10
cov.matrix[[7]][2,1] <- cov.matrix[[7]][1,2] <- 0.6 # corr = 0.67
cov.matrix[[7]][3,1] <- cov.matrix[[7]][1,3] <- 0.5
cov.matrix[[7]][3,2] <- cov.matrix[[7]][2,3] <- 0.4

## begin testing
test.sample.dep <- sample.dep( n.dep , n, var.ratio, cov.matrix, ngenes, delta, distrn = "normal")

## Not run:  

ndim <- dim(test.sample.dep);
var.r <- rep(0, ndim[1])

for ( i in c(1:ndim[1]))
  {
     var.r[i] <- var(test.sample.dep[i,(n+1):(2*n)]) /
                var(test.sample.dep[i,1:n])
  }

generated.cov <- cov(t(test.sample.dep[,1:n]), t(test.sample.dep[,1:n]))
# the fn cov compute the covariance between the columns

cat("\n comparison between the desired sample and the generated sample: \n")

cat(" ------------------------------------------ \n",
    "number of genes (rows) :\t", n.dep, "\t", ndim[1], "\n",
    "number of obs for each group: \t", n, "\t", ndim[2]/2, "\n",
    "variance ratio (trt/ctrl) is: \t", var.ratio, "\t", mean(var.r), "\n",
    "covariance matrix should be one of the following: \n")
# print(cov.matrix)

cat("the covariance matrix for the generated sample is: \n")
print(generated.cov)
cat("\n------------------------------------------ \n")


## End(Not run) 

warnes/exp.ssize documentation built on May 4, 2019, 12:59 a.m.