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

In StoreyLab/biobroom: Turn Bioconductor objects into tidy data frames

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

## S3 method for class 'MArrayLM'
tidy(x, intercept = FALSE, ...)

## S3 method for class 'MArrayLM'
augment(x, data, ...)

## S3 method for class 'MArrayLM'
glance(x, ...)

## S3 method for class 'MAList'
tidy(x, ...)

## S3 method for class 'EList'
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.

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).

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

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)
    }
}

StoreyLab/biobroom documentation built on May 9, 2019, 3:09 p.m.