\name{dyebias.estimate.iGSDBs}
\alias{dyebias.estimate.iGSDBs}
\title{Estimate intrinsic gene specific dye biases (part of the GASSCO method)}
\description{
Obtain estimates for the instrinsic gene-specific dye bias (iGSDB)
using a set of normalized data, as part of the GASSCO method.
}
\usage{
dyebias.estimate.iGSDBs(data.norm, is.balanced=TRUE, reference="ref",
verbose=FALSE)
}
\arguments{
\item{data.norm}{
A \code{marrayNorm} object containing the data for estimating the
dye bias. This object is supposed to be complete. In particular, \cr
\code{maLabels(maGnames(data.norm))} must be set and must indicate the
identities of the reporter sequence (i.e., oligo or cDNA
sequence) of each spot. This helps identify replicate spots, which
are averaged as part of the estimation.
If the data is unbalanced (so \code{is.balanced} is \code{FALSE}), \cr
\code{maInfo(maTargets(data.norm))} is also required, and should
contain at least two attributes: \code{Cy5} and \code{Cy3}. Both
should indicate the factor value for the respective channel.}
\item{is.balanced}{
The use of this argument is discouraged, since designs should generally be
balanced. The values other than \code{TRUE} will become illegal in
the future.
Logical indicating whether the data set represents a balanced design
(which is by far the most common case). A design is balanced if all
factor values are present an equal number of times in both the
forward and reverse dye orientations. A self-self design is by
definition balanced (even if the number of slides is uneven). If
\code{is.balanced} is \code{TRUE}, the iGSDB estimate is
obtained by simply averaging, per reporter, all \eqn{M} values
(and the value of the \code{reference} argument is ignored).
If \code{is.balanced==FALSE}, the design is inferred from the
\code{reference} argument, and subsequently the \code{limma}
package is used to model the dye effect. This is typically done for
an unbalanced data set, but there is no harm in setting
\code{is.balanced=FALSE} for a design that by itself is already
balanced. If there are no missing values in the data, the results of
using the simple average and the limma procedure are identical
(although LIMMA takes longer to compute the iGSDBs). If the data set
contains many missing data points (NA's), the limma estimates differ
slightly from the simple averaged estimates (although it is not
clear which ones are better).
}
\item{reference}{ If the design contains a single common reference,
the \code{reference} argument should be this common reference (which may not be
empty). If the design
contains multiple common references, \code{reference}
should be a vector listing all the common references, and the name
of the factor value that is not the common reference should have its
own common reference as a prefix. E.g., if two mutant strains
\code{mutA} and \code{mutB} were assayed, each against a separate
common reference \code{ref1} and \code{ref2}, the
\code{reference}-argument
would be \code{c("ref1", "ref2")}, and the \code{Cy3} and \code{Cy5}
attributes of \code{maInfo(maTargets(data.norm))} would be
values from \code{"ref1:mutA", "ref2:mutA", "ref1:mutB",
"ref2:mutB"}. The colon is not important, but the prefix is, as it
allows the association of each sample with its 'own' common reference.
}
\item{verbose}{
Logical, indicating wether or not to be verbose.
}
}
\value{
A data frame is returned with as many rows as there are reporters
(replicate spots have been averaged), and the following columns: \cr
\item{reporterId}{The name of the reporter}
\item{dyebias}{The intrinsic gene-specific dye bias (iGSDB) of this reporter}
\item{A}{The average expression level of this reporter in the given data set}
\item{p.value}{The \emph{p}-value for the \code{dyebias} ($H_0$:
\code{dyebias} = 0). All \emph{p}-value are set to NA if
they were not estimated (i.e., if limma was not run because
\code{is.balanced} was TRUE)}
This data frame is typically used as input to \code{\link{dyebias.apply.correction}}.
}
\examples{
\dontshow{
options(stringsAsFactors = FALSE)
library(dyebias)
library(dyebiasexamples)
data(data.raw)
data(data.norm)
} % dontshow
iGSDBs.estimated <- dyebias.estimate.iGSDBs(data.norm,
is.balanced=TRUE,
verbose=FALSE)
summary(iGSDBs.estimated)
\dontrun{
hist(iGSDBs.estimated$dyebias, breaks=50)
}
}
\details{
This function implements the first step of the GASSCO method:
estimating the so-called intrinsic gene specific dye biases, or
briefly iGSDB. They can be estimated from a (preferably large) data
set containing either self-self experiments, or dye-swapped slides.
The assumption underlying this approach is that with self-selfs, or
with pairs of dye swaps, the only effect that can lead to systematic
changes between Cy5 and Cy3, is in fact the dye effect.
There are two cases to distinguish, the balanced case, and the
unbalanced case. In the balanced case, the iGSDB estimate is simply
the average \eqn{M} (where \cr \eqn{M = log_2(R/G) = log_2(Cy5/Cy3)}) over all
slides. A set of slides is balanced if all factor values are present
in as many dye-swapped as non-dye-swapped slides. A set of self-self
slides is in fact a degenerate form of this, and is therefore also
balanced.
In the unbalanced case, one could omit slides until the data set is
balanced. However, this is wasteful as we can use linear modelling to
obtain estimates. We use the limma package for this (Smyth, 2005). The only
unbalanced designs currently supported are a common reference design,
and a set of common reference designs.
There are no weights or subset argument to this function; the
estimation is done for all reporters found. If there are replicate
spots, they are averaged prior to the estimation (the reason being
that we are not interested in p-values for the estimate)
Having obtained the iGSDB estimates, the corrections can be applied
to either to the hybridizations given by the \code{data.norm} argument,
or to a different set of slides that is thought to have very similar
iGSDBs. Applying the corrections is done with \cr
\code{\link{dyebias.apply.correction}}.
}
\seealso{dyebias.apply.correction}
\note{
Note that the input data should be normalized, and that the dye swaps
should \strong{not} have been swapped back. After all, we're
interested in the difference of Cy5 over Cy3, \strong{not} the
difference of experiment over reference.
}
\author{Philip Lijnzaad \email{p.lijnzaad@umcutrecht.nl}}
\references{
Margaritis, T., Lijnzaad, P., van Leenen, D., Bouwmeester, D.,
Kemmeren, P., van Hooff, S.R and Holstege, F.C.P. (2009)
Adaptable gene-specific dye bias correction for two-channel DNA microarrays.
\emph{Molecular Systems Biology}, 5:266, 2009. doi: 10.1038/msb.2009.21.
Dudoit, S. and Yang, Y.H. (2002) Bioconductor R packages for
exploratory analysis and normalization of cDNA microarray data. In:
Parmigiani, G., Garrett, E.S. , Irizarry, R.A., and Zeger, S.L. (eds.)
\emph{The Analysis of Gene Expression Data: Methods and Software},
Springer, New York.
%% Smyth, G.K. (2004) Linear models and empirical Bayes methods for
%% assessing differential expression in microarray
%% experiments. \emph{Statistical Applications in Genetics and Molecular
%% Biology}, \bold{3}, article 3.
Smyth, G.K. (2005) Limma: linear models for microarray data.
In: Gentleman, R., Carey, V., Dudoit, S., Irizarry, R. and
Huber, W. (eds). \emph{Bioinformatics and Computational Biology
Solutions using R and Bioconductor}, Springer, New York.
}
% Add one or more standard keywords, see file 'KEYWORDS' in the
% R documentation directory.
\keyword{misc} % silly, but one keyword is compulsary
