tranestAffyProbeLevel: Glog transformation parameter estimation function for...
In LMGene: LMGene Software for Data Transformation and Identification of Differentially Expressed Genes in Gene Expression Arrays

Description Usage Arguments Details Value Author(s) References See Also Examples

Estimates parameters for the glog transformation on probe-level Affymetrix expression data, by maximum likelihood or by minimizing the stability score.

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)

`eS`	An `AffyBatch` object
`ngenes`	Number of randomly sampled probesets to be used in estimating the transformation parameter
`starting`	If `TRUE`, user-specified starting values for `lambda` and `alpha` are input to the optimization routine
`lambda`	Starting value for parameter `lambda`. Ignored unless `starting = TRUE`
`alpha`	Starting value for parameter `alpha`. Ignored unless `starting = TRUE`
`gradtol`	A positive scalar giving the tolerance at which the scaled gradient is considered close enough to zero to terminate the algorithm
`lowessnorm`	If `TRUE`, lowess normalization (using `lnorm`) is used in calculating the likelihood.
`method`	Determines optimization method. Default is 1, which corresponds to a Newton-type method (see `nlm` and details.)
`mult`	If `TRUE`, tranest will use a vector alpha with one (possibly different) entry per sample. Default is to use same alpha for every sample. `SD` and `mult` may not both be `TRUE`.
`model`	Specifies model to be used. Default is to use all variables from eS without interactions. See details.
`SD`	If `TRUE`, transformation parameters are estimated by minimizing the stability score. See details.
`rank`	If `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 `SD = TRUE`
`model.based`	If `TRUE`, the stability score is calculated using the standard deviation of residuals from the linear model in `model`. Ignored unless `SD = TRUE`
`rep.arrays`	List of sets of replicate arrays. Each element of `rep.arrays` should be a vector with entries corresponding to arrays (columns) in `exprs(eS)` conducted under the same experimental conditions, i.e., with identical rows in `pData(eS)`. Ignored unless `SD = TRUE` and `model.based = FALSE`

The model argument is an optional character string, constructed like the right-hand side of a formula for lm. It specifies which of the variables in the ExpressionSet will be used in the model and whether interaction terms will be included. If model=NULL, it uses all variables from the 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 lambda and alpha such that the residuals from the linear model in model, fit to glog-transformed data using those values for lambda and alpha, follow a normal distribution. See Durbin and Rocke (2003) for details.

If SD = TRUE, lambda and 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 model.based = TRUE, the stability score is calculated using the standard deviation of residuals from the linear model in model. Otherwise, the stability score is calculated using the pooled standard deviation over sets of replicates in rep.arrays. See Wu and Rocke (2009) for details.

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

Optimization methods in method are as follows:

1 =: Newton-type method, using nlm
2 =: Nelder-Mead, using optim
3 =: BFGS, using optim
4 =: Conjugate gradients, using optim
5 =: Simulated annealing, using optim (may only be used when mult = TRUE)

A list with components:

`lambda`	Estimate of transformation parameter lambda
`alpha`	Estimate of transformation parameter alpha

Lei Zhou, David Rocke, Geun-Cheol Lee, John Tillinghast, Blythe Durbin-Johnson, and Shiquan Wu

Durbin, B.P and Rocke, D.M. (2003) Estimation of Transformation Parameters for Microarray Data, Bioinformatics, 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, Bioinformatics, 21, 3983–3989.

http://dmrocke.ucdavis.edu

tranest, lnorm, psmeans, glog

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)
## Not run: 
# 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)

## End(Not run)