Compute row-wise mean and variance using matrix operations

Description

In large data sets, like those arising from microarray or proteomics experiments, it is often necessary to compute the mean and variance for each row among many thousands. The computations can be significantly faster if vectorized instead of using the inherent loop in an apply statement.

Usage

matrixMean(x)
matrixVar(x, xmean)

Arguments

Value

A vector containing a number of entries equal to the number of rows of x, where each entry is either the mean or the variance of the corresponding row.

Author(s)

Kevin R. Coombes <kcoombes@mdanderson.org>

See Also

apply

Examples

nr <- 10000
nc <- 40
fake.data <- matrix(rnorm(nr*nc), nrow=nr)
fake.class <- rep(c('H', 'C'), each=20)

H.n <- sum(fake.class=='H')
C.n <- sum(fake.class=='C')

H.mean <- matrixMean(fake.data[, fake.class=='H'])
H.var <- matrixVar(fake.data[, fake.class=='H'], H.mean)

C.mean <- matrixMean(fake.data[, fake.class=='C'])
C.var <- matrixVar(fake.data[, fake.class=='C'], C.mean)

pooled.sd <- sqrt( (H.var*(H.n - 1) + C.var*(C.n - 1))/(H.n + C.n - 2) ) 

t.statistics <- (C.mean - H.mean)/pooled.sd/sqrt(1/H.n + 1/C.n)