glmFit: Genewise Negative Binomial Generalized Linear Models

Description

Fit a negative binomial generalized log-linear model to the read counts for each gene. Conduct genewise statistical tests for a given coefficient or coefficient contrast.

Usage

"glmFit"(y, design=NULL, dispersion=NULL, prior.count=0.125, start=NULL, ...)
"glmFit"(y, design=NULL, dispersion=NULL, offset=NULL, lib.size=NULL, weights=NULL, prior.count=0.125, start=NULL, ...)
glmLRT(glmfit, coef=ncol(glmfit$design), contrast=NULL)

Arguments

an object that contains the raw counts for each library (the measure of expression level); alternatively, a matrix of counts, or a DGEList object with (at least) elements counts (table of unadjusted counts) and samples (data frame containing information about experimental group, library size and normalization factor for the library size)

design

numeric matrix giving the design matrix for the genewise linear models. Must be of full column rank. Defaults to a single column of ones, equivalent to treating the columns as replicate libraries.

dispersion

numeric scalar, vector or matrix of negative binomial dispersions. Can be a common value for all genes, a vector of dispersion values with one for each gene, or a matrix of dispersion values with one for each observation. If NULL will be extracted from y, with order of precedence: genewise dispersion, trended dispersions, common dispersion.

offset

numeric matrix of same size as y giving offsets for the log-linear models. Can be a scalor or a vector of length ncol(y), in which case it is expanded out to a matrix.

weights

optional numeric matrix giving prior weights for the observations (for each library and gene) to be used in the GLM calculations.

lib.size

numeric vector of length ncol(y) giving library sizes. Only used if offset=NULL, in which case offset is set to log(lib.size). Defaults to colSums(y).

prior.count

average prior count to be added to observation to shrink the estimated log-fold-changes towards zero.

start

optional numeric matrix of initial estimates for the linear model coefficients.

...

other arguments are passed to lower level fitting functions.

glmfit

a DGEGLM object, usually output from glmFit.

coef

integer or character vector indicating which coefficients of the linear model are to be tested equal to zero. Values must be columns or column names of design. Defaults to the last coefficient. Ignored if contrast is specified.

contrast

numeric vector or matrix specifying one or more contrasts of the linear model coefficients to be tested equal to zero. Number of rows must equal to the number of columns of design. If specified, then takes precedence over coef.

Value

design: design matrix as input.
weights: matrix of weights as input.
df.residual: numeric vector of residual degrees of freedom, one for each gene.
offset: numeric matrix of linear model offsets.
dispersion: vector of dispersions used for the fit.
coefficients: numeric matrix of estimated coefficients from the glm fits, on the natural log scale, of size nrow(y) by ncol(design).
unshrunk.coefficients: numeric matrix of estimated coefficients from the glm fits when no log-fold-changes shrinkage is applied, on the natural log scale, of size nrow(y) by ncol(design). It exists only when prior.count is not 0.
fitted.values: matrix of fitted values from glm fits, same number of rows and columns as y.
deviance: numeric vector of deviances, one for each gene.
table: data frame with the same rows as y containing the log2-fold-changes, likelhood ratio statistics and p-values, ready to be displayed by topTags.
comparison: character string describing the coefficient or the contrast being tested.
logFC: log2-fold change of expression between conditions being tested.
logCPM: average log2-counts per million, the average taken over all libraries in y.
LR: likelihood ratio statistics.
PValue: p-values.

Details

glmFit and glmLRT implement generalized linear model (glm) methods developed by McCarthy et al (2012).

glmFit fits genewise negative binomial glms, all with the same design matrix but possibly different dispersions, offsets and weights. When the design matrix defines a one-way layout, or can be re-parametrized to a one-way layout, the glms are fitting very quickly using mglmOneGroup. Otherwise the default fitting method, implemented in mglmLevenberg, uses a Fisher scoring algorithm with Levenberg-style damping.

Positive prior.count cause the returned coefficients to be shrunk in such a way that fold-changes between the treatment conditions are decreased. In particular, infinite fold-changes are avoided. Larger values cause more shrinkage. The returned coefficients are affected but not the likelihood ratio tests or p-values.

glmLRT conducts likelihood ratio tests for one or more coefficients in the linear model. If coef is used, the null hypothesis is that all the coefficients indicated by coef are equal to zero. If contrast is non-null, then the null hypothesis is that the specified contrasts of the coefficients are equal to zero. For example, a contrast of c(0,1,-1), assuming there are three coefficients, would test the hypothesis that the second and third coefficients are equal.

References

McCarthy, DJ, Chen, Y, Smyth, GK (2012). Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation. Nucleic Acids Research 40, 4288-4297. http://nar.oxfordjournals.org/content/40/10/4288

Examples

Run this code

nlibs <- 3
ngenes <- 100
dispersion.true <- 0.1

# Make first gene respond to covariate x
x <- 0:2
design <- model.matrix(~x)
beta.true <- cbind(Beta1=2,Beta2=c(2,rep(0,ngenes-1)))
mu.true <- 2^(beta.true %*% t(design))

# Generate count data
y <- rnbinom(ngenes*nlibs,mu=mu.true,size=1/dispersion.true)
y <- matrix(y,ngenes,nlibs)
colnames(y) <- c("x0","x1","x2")
rownames(y) <- paste("gene",1:ngenes,sep=".")
d <- DGEList(y)

# Normalize
d <- calcNormFactors(d)

# Fit the NB GLMs
fit <- glmFit(d, design, dispersion=dispersion.true)

# Likelihood ratio tests for trend
results <- glmLRT(fit, coef=2)
topTags(results)

# Estimate the dispersion (may be unreliable with so few genes)
d <- estimateGLMCommonDisp(d, design, verbose=TRUE)

Run the code above in your browser using DataLab