\name{tranestAffyProbeLevel}
\alias{tranestAffyProbeLevel}
\title{
Glog transformation parameter estimation function for probe-level Affymetrix expression data
}
\description{
Estimates parameters for the glog transformation on probe-level Affymetrix expression data, by maximum likelihood 
or by minimizing the stability score.
}
\usage{
tranestAffyProbeLevel(eS, ngenes = 5000, starting = FALSE, lambda = 1000, 
alpha = 0, gradtol = 0.001,lowessnorm = FALSE, method = 1, mult = FALSE, 
model = NULL, SD = FALSE, rank = TRUE, model.based = TRUE, 
rep.arrays = NULL)
}
\arguments{
  \item{eS}{An \code{AffyBatch} object}
  \item{ngenes}{Number of randomly sampled probesets to be used in estimating the transformation parameter}
  \item{starting}{If \code{TRUE}, user-specified starting values for \code{lambda} and \code{alpha} are input to 
the optimization routine}
  \item{lambda}{Starting value for parameter \code{lambda}. Ignored unless \code{starting = TRUE}}
  \item{alpha}{Starting value for parameter \code{alpha}. Ignored unless \code{starting = TRUE}}
  \item{gradtol}{A positive scalar giving the tolerance at which the scaled 
gradient is considered close enough to zero to terminate the algorithm
}
  \item{lowessnorm}{If \code{TRUE}, lowess normalization (using \code{\link{lnorm}}) is used in calculating 
the likelihood.}
  \item{method}{Determines optimization method. Default is 1, 
which corresponds to a Newton-type method (see \code{nlm} and details.)}
  \item{mult}{If \code{TRUE}, tranest will use a vector alpha with one (possibly different) entry per sample. 
Default is to use same alpha for every sample.  \code{SD} and \code{mult} may not both be \code{TRUE}.}
  \item{model}{Specifies model to be used. Default is to use all variables from eS without interactions. See details.}
  \item{SD}{If \code{TRUE}, transformation parameters are estimated by minimizing the stability score.  See details.}
  \item{rank}{If \code{TRUE}, the stability score is calculated by regressing the replicate standard deviation
on the rank of the probe/row means (rather than on the means themselves).  Ignored unless \code{SD = TRUE}}
  \item{model.based}{If \code{TRUE}, the stability score is calculated using the standard deviation of residuals from the linear
    model in \code{model}.  Ignored unless \code{SD = TRUE}}
  \item{rep.arrays}{List of sets of replicate arrays.  Each element of \code{rep.arrays} should be a vector with entries
corresponding to arrays (columns) in \code{exprs(eS)} conducted under the same experimental conditions, i.e., with identical
rows in \code{pData(eS)}. Ignored unless \code{SD = TRUE} and \code{model.based = FALSE}}
}
\details{
  The \code{model} argument is an optional character string, constructed like the right-hand
  side of a formula for \code{lm}. It specifies which of the variables in the \code{ExpressionSet} will
  be used in the model and whether interaction terms will be included. If \code{model=NULL},
  it uses all variables from the \code{ExpressionSet} without interactions. Be careful of using
  interaction terms with factors; this often leads to overfitting, which will yield an error.

  The default estimation method is maximum likelihood.  The likelihood is derived by assuming that there exist values for \code{lambda}
  and \code{alpha} such that the residuals from the linear model in \code{model}, fit to glog-transformed data using those values 
  for \code{lambda} and \code{alpha}, follow a normal distribution.  See Durbin and Rocke (2003) for details.

  If \code{SD = TRUE}, \code{lambda} and \code{alpha} are estimated by minimizing the stability score rather than by maximum likelihood.
  The stability score is defined as the absolute value of the slope coefficient from the regression of the replicate/residual 
  standard deviation on the probe/row means, or on the rank of the probe/row means.  If \code{model.based = TRUE}, the stability
  score is calculated using the standard deviation of residuals from the linear model in \code{model}.  Otherwise, the stability score is 
  calculated using the pooled standard deviation over sets of replicates in \code{rep.arrays}. See Wu and Rocke (2009) for details.

  A random sample of probsets (of size \code{ngene}) is sampled from \code{featureNames(eS)}.  Expression data from all probes in the sampled 
  probesets is used in estimating the transformation parameters. 

Optimization methods in \code{method} are as follows:
\describe{
\item{1 = }{Newton-type method, using \code{nlm}}
\item{2 = }{Nelder-Mead, using \code{optim}}
\item{3 = }{BFGS, using \code{optim}}
\item{4 = }{Conjugate gradients, using \code{optim}}
\item{5 = }{Simulated annealing, using \code{optim} (may only be used when \code{mult = TRUE})}
}
}
\value{
A list with components: 
\item{lambda}{Estimate of transformation parameter lambda}
\item{alpha}{Estimate of transformation parameter alpha}
}
\references{

Durbin, B.P and Rocke, D.M. (2003) Estimation of Transformation Parameters for Microarray Data,  
\emph{Bioinformatics}, \bold{19}, 1360--1367.

Wu, S. and Rocke, D.M. (2009) Analysis of Illumina BeadArray data using variance stabilizing transformations.

Zhou, L. and Rocke, D.M. (2005) An expression index for Affymetrix GeneChips based on the generalized logarithm,
\emph{Bioinformatics}, \bold{21}, 3983--3989.

\url{http://dmrocke.ucdavis.edu}
}
\author{
Lei Zhou, David Rocke, Geun-Cheol Lee, John Tillinghast, Blythe Durbin-Johnson, and Shiquan Wu
}
\seealso{
\code{\link{tranest}}, \code{\link{lnorm}}, \code{\link{psmeans}}, \code{\link{glog}}}
\examples{
library(LMGene)
library(affy)
library(Biobase)
library(affydata)

data(Dilution) 

tranpar.Dilution <- tranestAffyProbeLevel(Dilution, model = "liver", 
ngenes = 3000, method = 2)

# transform data
trans.Dilution <- transeS(Dilution, tranpar.Dilution$lambda,
		tranpar.Dilution$alpha)

# extract transformed perfect matches
exprs(trans.Dilution) <- pm(trans.Dilution)

# lowess normalize transformed data
lnorm.Dilution <- lnormeS(trans.Dilution)
\dontrun{
# Average over probesets
# First, create index of probes
fnames <- featureNames(Dilution)
p <- length(featureNames(Dilution))
ind <- vector() 
for (i in 1:p){
	nprobes <- dim(pm(Dilution,fnames[i]))[1]
	ind <- c(ind, rep(i,nprobes))   
}

avg.Dilution <- psmeans(lnorm.Dilution, ind)
}
}
% Add one or more standard keywords, see file 'KEYWORDS' in the
% R documentation directory.
\keyword{ math}