sigaR (version 1.18.0)

RCMtest: Hypothesis testing within the random coefficient model.

Description

Function that evaluates various hypothesis within the random coefficients model via bootstrap resampling.

Usage

RCMtest(Y, X, R, testType = "I", nBoot = 100, lowCiThres = 0.1, 
shrinkType = "none", estType = "normal", corType = "unif", 
maxNoIt = 100, minSuccDist = 0.005, returnNullDist = FALSE, 
ncpus=1, verbose = FALSE)

Arguments

Y

The matrix containing the (e.g., expression) data (number of columns equal to number of features, number of rows equal to number of samples).

X

The design matrix (number of rows equal to number of samples, number of columns equal to number of covariates).

R

The linear constraint matrix (number of columns equal to the number of covariates).

testType

The hypothesis to be tested: I (H0 : R beta = 0 & tau2 = 0) vs. (H2 : R beta >= 0 V tau2 >= 0), II (H0 : R beta = 0 & tau2 = 0) vs. (H1 : R beta >= 0 & tau2 = 0), III (H1 : R beta >= 0 & tau2 = 0) vs. (H2 : R beta >= 0 & tau2 >= 0).

nBoot

Number of bootstraps.

lowCiThres

A value between 0 and 1. Determines speed of efficient p-value calculation. If the probability of a p-value being below lowCiThres is smaller than 0.001 (read: the test is unlikely to become significant), bootstrapping is terminated and a p-value of 1.00 is reported.

shrinkType

The type of shrinkage to be applied to the error variances: none (shrinkage parameter is set equal to zero: no shrinkage), opt (shrinkage parameter is chosen to minimize the mean squared error criterion) or full (shrinkage parameter is set equal to one).

estType

Type of estimation, either normal (non-robust) or robust.

corType

Correlation structure to be used, either unif or ar1.

maxNoIt

Maximum number of iterations in the ML procedure.

minSuccDist

Minimum distance between estimates of two successive iterations to be achieved.

returnNullDist

Logical indicator: should the null distribution be returned?

ncpus

Number of cpus used for the bootstrap.

verbose

Logical indicator: should intermediate output be printed on the screen?

Value

Object of class rcmTest.

Warning

In case a covariate for the intercept is included in the design matrix X we strongly recommend the center, per feature, the data around zero.

Details

Details on the type of random coefficients model that is actually fitted are specified in the reference below.

References

Van Wieringen, W.N., Berkhof, J., Van de Wiel, M.A. (2010), "A random coefficients model for regional co-expression associated with DNA copy number", Statistical Applications in Genetics and Molecular Biology, Volume 9, Issue1, Article 25, 1-28.

Van Wieringen, W.N., Van de Wiel, M.A., Van der Vaart, A.W. (2008), "A test for partial differential expression", Journal of the American Statistical Association, 103(483), 1039-1049.

See Also

RCMestimation, RCMrandom, rcmTest.

Examples

# load data
data(pollackCN16)
data(pollackGE16)

# select features belonging to a region
ids <- getSegFeatures(20, pollackCN16)

# extract segmented log2 ratios of the region
X <- t(segmented(pollackCN16)[ids[1], , drop=FALSE])

# extract segmented log2 ratios of the region
Y <- exprs(pollackGE16)[ids,]

# center the expression data (row-wise)
Y <- t(Y - apply(Y, 1, mean))

# specify the linear constraint matrix
R <- matrix(1, nrow=1)

# fit the random coefficients model to the random data
RCMresults <- RCMestimation(Y, X, R)

# test for significance of effect of X on Y
RCMtestResults <- RCMtest(Y, X, R, nBoot=2) 
summary(RCMtestResults)