TailRank (version 3.2.4)

TailRankTest: The Tail-Rank Test

Description

Perform a tail-rank test to find candidate biomarkers in a microarray data set.

Usage

TailRankTest(data, classes, specificity = 0.95, tolerance = 0.50,
             model=c("bb", "betabinomial", "binomial"),
             confidence = 0.95, direction = "up")

Value

The return value is an object of class

TailRankTest.

Arguments

data

A matrix or data.frame containing numerical measurements on which to perform the tail-rank test.

classes

A logical vector or factor splitting the data into two parts. The length of this vector should equal the number of columns in the data. The TRUE portion (or the first level of the factor) represents a "base" or "healthy" group of samples; the other samples are the "test" or "cancer" group.

specificity

a real number between 0 and 1; the desired specificity used in the test to estimate a quantile from the "base" group. This is an optional argument with default value 0.95.

tolerance

a real number between 0 and 1; the upper tolerance bound used to estimate the threshold. This is an optional argument with default value 0.90.

model

a character string that determines whther significance comes from a binomial model or a beta-binomial (bb) model.

confidence

a real number between 0 and 1; the confidence level that there are no false positives. This is an optional argument with default value 0.50, which is equivalent to ignoring the tolerance.

direction

a character string representing the direction of the test; can be "up", "down", or "two-sided". The default value is "up".

Author

Kevin R. Coombes <[e-mail]>

Details

This function computes the tail rank statistic for each gene (viewed as one row of the data matrix). The data is split into two groups. The first ("base") group is used to estimate a tolerance bound (defaults to 50%) on a specific quantile (defaults to 95%) of the distribution of each gene. The tail-rank statistic is the defined as the number of samples in the second ("test") group that lie outside the bound. The test can be applied in the "up", "down", or "two-sided" direction, depending on the kinds of markers being sought. Also computes the cutoff for significance based on a confidence level that is "1 - FWER" for a desired family-wise error rate.

References

http://bioinformatics.mdanderson.org

See Also

TailRankTest-class, tailRankPower, biomarkerPowerTable, toleranceBound

Examples

# generate some fake data to use in the example
nr <- 40000
nc <- 110
fake.data <- matrix(rnorm(nr*nc), ncol=nc)
fake.class <- rep(c(TRUE, FALSE), c(40, 70))

# perform the tail-rank test
null.tr <- TailRankTest(fake.data, fake.class)

# get a summary of the results
summary(null.tr)

# plot a histogram of the statistics
hist(null.tr, overlay=TRUE)

# get the actual statistics
stats <- getStatistic(null.tr)

# get a vector that selects the "positive" calls for the test
is.marker <- as.logical(null.tr)

# the following line should evaluate to the number of rows, nr = 40000
sum( is.marker == (stats > null.tr@cutoff) )