biobroom (version 1.4.2)

limma_tidiers: Tidiers for the output of limma (linear models for microarray analysis)

Description

Tidy, augment, and glance methods for MArrayLM objects, which contain the results of gene-wise linear models to microarray datasets. This class is the output of the lmFit and eBayes functions.

Tidying method for a MA list

Tidy an EList expression object

Usage

"tidy"(x, intercept = FALSE, ...)
"augment"(x, data, ...)
"glance"(x, ...)
"tidy"(x, ...)
"tidy"(x, addTargets = FALSE, ...)

Arguments

x

MArrayLM, MAList, Elist object

intercept

whether the (Intercept) term should be included (default FALSE)

...

extra arguments, not used

data

original expression matrix; if missing, augment returns only the computed per-gene statistics

addTargets

Add sample level information. Default is FALSE.

Value

The output of tidying functions is always a data frame without rownames.tidy returns one row per gene per coefficient. It always contains the columns

gene

The name of the gene (extracted from the rownames of the input matrix)

term

The coefficient being estimated

estimate

The estimate of each per-gene coefficient

Depending on whether the object comes from eBayes, it may also contain

statistic

Empirical Bayes t-statistic

p.value

p-value computed from t-statistic

lod

log-of-odds score

augment returns one row per gene, containing the original gene expression matrix if provided. It then adds columns containing the per-gene statistics included in the MArrayLM object, each prepended with a .:

.gene

gene ID, obtained from the rownames of the input

.sigma

per-gene residual standard deviation

.df.residual

per-gene residual degrees of freedom

The following columns may also be included, depending on which have been added by lmFit and eBayes:

.AMean

average intensity across probes

.statistic

moderated F-statistic

.p.value

p-value generated from moderated F-statistic

.df.total

total degrees of freedom per gene

.df.residual

residual degrees of freedom per gene

.s2.post

posterior estimate of residual variance

glance returns one row, containing

rank

rank of design matrix

df.prior

empirical Bayesian prior degrees of freedom

s2.prior

empirical Bayesian prior residual standard deviation

tidy returns a data frame with one row per gene-sample combination, with columns

gene

gene name

sample

sample name (from column names)

value

expressions on log2 scale

tidy returns a data frame with one row per gene-sample combination, with columns

gene

gene name

sample

sample name (from column names)

value

expressions on log2 scale

weight

present if weights is set

other columns

if present and if addTargets is set

Details

Tidying this fit computes one row per coefficient per gene, while augmenting returns one row per gene, with per-gene statistics included. (This is thus a rare case where the augment output has more rows than the tidy output. This is a side effect of the fact that the input to limma is not tidy but rather a one-row-per-gene matrix).

Examples

if (require("limma")) {
    # create random data and design
    set.seed(2014)
    dat <- matrix(rnorm(1000), ncol=4)
    dat[, 1:2] <- dat[, 1:2] + .5  # add an effect
    rownames(dat) <- paste0("g", 1:nrow(dat))
    des <- data.frame(treatment = c("a", "a", "b", "b"),
                      confounding = rnorm(4))

    lfit <- lmFit(dat, model.matrix(~ treatment + confounding, des))
    eb <- eBayes(lfit)
    head(tidy(lfit))
    head(tidy(eb))

    if (require("ggplot2")) {
        # the tidied form puts it in an ideal form for plotting
        ggplot(tidy(lfit), aes(estimate)) + geom_histogram(binwidth=1) +
            facet_wrap(~ term)
        ggplot(tidy(eb), aes(p.value)) + geom_histogram(binwidth=.2) +
            facet_wrap(~ term)
    }
}