robust.g.test {GeneTS}R Documentation

Robust g Test for Multiple (Genetic) Time Series

Description

robust.g.test calculates the p-value(s) for a robust nonparametric version of Fisher's g-test (1929). Details of this approach are described in Ahdesmaki et al. (2005), along with an extensive discussion of its application to gene expression data.

g.statistic computes the test statistic given a discrete time series spectrum.

robust.spectrum computes a robust rank-based estimate of the periodogram/correlogram - see Ahdesmaki et al. (2005) for details.

Usage

robust.g.test(y, index, perm = FALSE, x, noOfPermutations = 5000)
g.statistic(y, index)
robust.spectrum(x)

Arguments

y the matrix consisting of the spectral estimates as column vectors
x a matrix consisting of the time series as column vectors. In robust.g.test only needed if permutation tests are used
index an index to the spectral estimates that is to be used in the testing for periodicity. If index is missing, the maximum component of the spectral estimate is used in testing (regardless of the frequency of this maximum)
perm if perm is FALSE, a simulated distribution for the g-statistic is used. If per perm is TRUE, permutation tests are used to find the distribution of the g-statistic for each time series separately.
noOfPermutations number of permutations that are used for each time series (default = 5000)

Details

Application of robust.g.test can be very computer intensive, especially the production of the distribution of the test statistics may take a lot of time. Therefore, this distribution (dependening on the length of the time series) is stored in an external file to avoid recomputation (see example below).

For the general idea behind the Fisher's g test also see fisher.g.test which implements an analytic approach for g-testing. This is faster but not robust and also assumes Gaussian noise.

Value

robust.g.test returns a list of p-values, and g.statistic the associated test statistics. robust.spectrum returns a matrix where the column vectors correspond to the spectra corresponding to each time series.

Author(s)

Miika Ahdesmaki (miika.ahdesmaki@tut.fi).

References

Fisher, R.A. (1929). Tests of significance in harmonic analysis. Proc. Roy. Soc. A, 125, 54–59.

Ahdesmaki, M., Lahdesmaki, H., Peason, R., Huttunen, H., and Yli-Harja O. (2005). BMC Bioinformatics 6:117.

See Also

fdr.control, fisher.g.test.

Examples

## Not run: 

# load GeneTS library
library("GeneTS")

# load data set
data(caulobacter)

# how many samples and and how many genes?
dim(caulobacter)

# robust, rank-based spectral estimator applied to first 5 genes
spe5 <- robust.spectrum(caulobacter[,1:5])

# g statistics computed from the spectrum
g.statistic(spe5)

# robust p-values
pval <- robust.g.test(spe5)  # generates a file with the name "g_pop_length_11.txt"
pval <- robust.g.test(spe5)  # second call: much faster..

pval

# delete the external file 
unlink("g_pop_length_11.txt") 

## End(Not run)

[Package GeneTS version 2.8.0 Index]