scale_rows: Centers and scales the rows of a numeric matrix.

In lianos/multiGSEA: A unified interface to a plethora of gene set enrichment analysis methods

Description

This was for two reasons: (1) to avoid the (more commonly used) t(scale(t(x), ...) idiom; and (2) to specify what values, columns of x, etc. to use to calculate means and sd's to use in the scaling function.

Usage

scale_rows(x, center = TRUE, scale = TRUE, ...)

Arguments

x	the matrix-like object
center	Either a logical, character, or numeric-like value that specifies what to center
scale	Either a logical, characeter, or numeric-like value that specifies what to scale

Details

For instance, you might want to subtract the mean of a subset of columns from each row in the matrix (like the columns that come from control samples)

Value

a scaled version of x

Transformation based on specific columns

center and scale can be a logical, character, or numeric-like vector. The flexibility enables the following scenarios:

1. The user can set it to TRUE to center all values on the mean of their row. (FALSE does no centering)

2. A (named) vector of values that is a superset of rownames(x). These will be the values that are subtracted from each row.

3. A logical vector as long as ncol(x). Each value will be centered to the mean of the values of the columns specified as TRUE.

4. An integer vector, the is the analog of 3 but specifies the columns to use for centering.

Examples

# see tests/testthat/test-scale_rows.R for more examples
m <- matrix(rnorm(50, mean = 1, sd = 2), nrow = 5,
            dimnames = list(LETTERS[1:5], letters[1:10]))
s0 <- scale_rows(m, center = TRUE, scale = FALSE)
all.equal(s0, t(scale(t(m), center = TRUE, scale = FALSE)))

# mean center rows to a specific group of control samples (columns)
ctrl <- sample(colnames(m), 3)
s.ctrl <- scale_rows(m, center = ctrl, scale = FALSE)
ctrl.means <- Matrix::rowMeans(m[, ctrl])
all.equal(s.ctrl, t(scale(t(m), center = ctrl.means, scale = FALSE)))

lianos/multiGSEA documentation built on Nov. 17, 2020, 1:26 p.m.