globaltest {globaltest}R Documentation

Global Test

Description

In microarray data, tests a (list of) group(s) of genes for significant association with a given clinical variable.

Usage

globaltest(X, Y, genesets, model,
    levels, d, event = 1, adjust,
    method = c("auto", "asymptotic", "permutations", "gamma"),
    nperm = 10^4, scaleX = TRUE, accuracy = 50, ...) 

Arguments

X Either a matrix of gene expression data, where columns correspond to samples and rows to genes or a Bioconductor ExpressionSet. The data should be properly normalized beforehand (and log- or otherwise transformed), but missing values are allowed (coded as NA). Gene and sample names can be included as the row and column names of X.
Y A vector with the clinical outcome of interest, having one value for each sample. If X is an ExpressionSet it can also be the name of a covariate in the phenoData from the ExpressionSet, or a formula object using these names. If the clinical outcome is survival, Y should contain the survival times.
genesets Either a vector or a list of vectors. Indicates the group(s) of genes to be tested. Each vector in genesets can be given in three formats. Either it can be a vector with 1 (TRUE) or 0 (FALSE) for each gene in X, with 1 indicating that the gene belongs to the group. Or it can be a vector containing the column numbers (in X) of the genes belonging to the group. Or it can be a subset of the rownames or featureNames for X.
model Globaltest will try to determine the correct model from the input of Y and d. To override the automatic choice, use model = "logistic" for a two-valued outcome Y , model = "linear" for a continuous outcome and model = "survival" for a survival outcome.
levels If Y is a factor (or a category in the PhenoData slot of X) and contains more than 2 levels: levels is a vector of levels of Y to test. If levels is length 2: test these 2 groups against each other. If levels is length 1: test that level against the others.
d A vector or the name of a covariate in the phenoData from the ExpressionSet X, to indicate which samples experienced an event. Providing a value for d automatically sets model = "survival"
event The value or values of d that indicates that there was an event.
adjust Confounders or risk factors for which the test must be adjusted. Must be either a data frame or (if X is an ExpressionSet) the names of covariates in the phenoData from X or a formula object using these names. Default: no adjustment.
method The method for calculation the p-value. Use code{method = "asymptotic"} for the full asymptotic distribution of the test statistic; method = "gamma" for the gamma (= scaled chi-squared) approximation to that distribution and method = "permutations" for a permutation p-value. The default: method = "auto" chooses the permutations method if the number of possible permutations does not exceed 10,000 and the asymptotic otherwise. Note that method = "gamma" was the default option prior to version 4.0.0.
nperm A number of permutations. This gives the (maximum) number of permutations to be used if method = "permutations" or method = "auto".
scaleX If true, rescales the expression matrix to get pleasant value for all test statistics. The expression matrix X is multiplied by a constant in such a way that the expected value EQ of the test statistic for the global test becomes exactly 10. This rescaling has no effect on the p-values.
accuracy Numerical tuning parameter useable only with the asymptotic method and a non-survival response. Determines how much small eigenvalues of the R matrix are smoothed away to increase computation speed. Choose smaller values for quicker computations but conservative p-values; choose larger values for slower calculations but more accuracy.
... Captures deprecated input for compatibility with older versions of globaltest.

Details

The Global Test tests whether a group of genes (of any size from one single gene to all genes on the array) is significantly associated with a clinical variable. The group could be for example a known pathway, an area on the genome or the set of all genes. The test investigates whether samples with similar clinical outcomes tend to have similar gene expression patterns. For a significant result it is not necessary that the genes in the group have similar expression patterns, only that many of them are correlated with the outcome.

Value

The function returns an object of class gt.result.

Note

The options globaltest options sampling and permutation have been replaced by separate functions from version 3.0. See sampling and permutations.

Author(s)

Jelle Goeman: j.j.goeman@lumc.nl; Jan Oosting

References

For references, type: citation("globaltest"). See also the vignette GlobalTest.pdf included with this package.

See Also

Many more examples in the vignette! geneplot, sampleplot, sampling, gt.multtest, permutations, checkerboard, regressionplot.

Examples

    # Breast cancer data (ExpressionSet) from the Netherlands Cancer
    # Institute with annotation:
    data(vandeVijver)
    data(annotation.vandeVijver)

    # Many possible calls. See the vignette for more examples and explanation.
    globaltest(vandeVijver, "StGallen")
    globaltest(vandeVijver, "StGallen", annotation.vandeVijver)
    globaltest(vandeVijver, "Surv(TIMEsurvival, EVENTdeath)", annotation.vandeVijver)
    globaltest(vandeVijver, StGallen ~ Posnodes + StGallen, annotation.vandeVijver)
    globaltest(vandeVijver, "StGallen", method = "p")

    # Store the test result
    # See help(gt.result) for more options
    gt <- globaltest(vandeVijver, "StGallen", annotation.vandeVijver)
    gt[1:2]
    sort(gt)
    p.value(gt)

    # Also with simple vector/matrix input
    X <- matrix(rnorm(3000), 100, 30)  # random expression data
    Y <- 1:30                          # a response variable
    pathway <- 1:40                    # a pathway

    globaltest(X, Y)
    globaltest(X, Y, pathway)

[Package globaltest version 4.12.0 Index]