preprocessBeadSet: Pre-processing of BeadSetIllumina objects

Description

Performs a sequence of pre-processing routines on objects of class "BeadSetIllumina"

Usage

setNormOptions(shearInf1 = TRUE, transf = "root",
    method = "medianAF",
    minSize = suggestSh(shearInf1)$minSize,
    prob = suggestSh(shearInf1)$prob,
    nBins = suggestSh(shearInf1)$nBins,
    dist = suggestTr(transf)$dist,
    pNorm = suggestTr(transf)$pNorm,
    nthRoot = suggestTr(transf)$nthRoot,
    offset = suggestTr(transf)$offset,
    scale = suggestNo(method)$scale,
    nSD = 3, breaks = 200)
plotPreprocessing(BSData, normInd,
    normOpts = setNormOptions(shearInf1 = !is.null(normInd)),
    plotArray = 1, ...)
preprocessBeadSet(BSData, normInd,
    normOpts = setNormOptions(shearInf1 = !is.null(normInd)))

Arguments

shearInf1

If TRUE, only the signal-containing channel of Infinium I beads are used to define the homozygote asymptotes for the affine transformation (rotation and shearing). This may be more accurate than using all beads, as the variation a

transf

Character string denoting transformation. One of none, log (base 2), or root (defined by nthRoot)

method

Character string denoting channel normalization method for each array. One of none, quantNorm, medianAF, or linPeak. For quantile normalization, the limma package i

minSize

The homozygote asymptotes are found by drawing a straight line through quantile points distributed in bins along each axis. Only bins containing more than minSize points are used

prob

Numeric probabiliy used in the quantile-function, defining the points through which the asymptotes are drawn

nBins

The number of bins into which to divide the points along each axis before the homozygote asymptotes are drawn

dist

Character string defining the distance measure used for polar coordinates transformation of the signal. One of manhattan, euclidean, or minkowski. See cart2

pNorm

See cart2pol

nthRoot

Numeric used together with transf="root"

offset

A numeric offset added to each channel before transformation. Values below zero are set to NA during log- or root-transformation

scale

Used with method="linPeak"

nSD

The background signal is estimated as nSD times the estimated standard measurement error (found from the the parameterised noise levels for each channel)

breaks

The parameterisation of noise levels is based on a histogram of each channel, where the numeric breaks defines the smoothing (number of bins). See hist

BSData

"BeadSetIllumina" object not previously pre-processed

normInd

Matrix with logical indexes to sub-bead pool for each bead-type. See getNormInd

normOpts

List output from setNormOptions

...

Further arguments to plotEstimatedNoise

plotArray

Numeric index to a single array to plot

Value

Output from setNormOptions is a list with pre-processing options
The function plotPreprocessing is used for its side effects Output from preprocessIllumina is a "BeadSetIllumina" object with pre-processed assayData entries. A column noiseIntensity is added to phenoData, this is the (parameterized) standard error times nSD

Details

Using setNormOptions, default pre-processing options are suggested, and any changes may be specified. The effects of different options are studied using plotPreprocessing for a number of arbitrary arrays. This produces four plots; i) raw data scatter, ii) scatter including the estimated asymptotes for the affine transformation (red/green) including the quantile points used (blue dots), iii) the noise levels for the red and green channel after transformation, parameterized signal superimposed, based on the non-signal channels of Infinium I beads, and iv) scatter after transformation including new axes (green) and estimated noise levels (red dots).

For the affine transformation, it is important that enough quantile points are included to get reliable asymptotes. If there are few blue dots in plot ii), decrease the minSize option or set shearInf1 to FALSE. If the grey lines in plot iii) are too coarse (too few points) to get a good noise-parameterisation, increase breaks. Note also how the noise levels are affected by different transformations.

Pay close regard to how the transformation affects the shapes of the clouds in plot iv). Ideally, three well defined clouds protrude from the estimated origin, corresponding to the homozygotes which fall on the estimated axes and the heterozygotes which fall 45 degrees in between. Imagine a rubber band stretched over the ends of the three clouds. If the rubber band is straight (no transformation), the manhattan (or 1-norm minkowski) distance is the best option for polar coordinates. If the three points fall on a circle, the euclidean (or 2-norm minkowski) distance is the best option. If the rubber band forms a shape intermediate between a circle and a square (e.g. 4th-root transformation), the 5-norm minkowski distance or similar may the best choice. The function preprocessBeadSet calls several pre-processing routines in sequence. First shearRawSignal performs the affine transformations, then getNoiseDistributions estimates the distributions of the noise for each channel. Next, transformChannels transforms the signal, followed by transformation of the standard errors of each channel using transformSEs. In the end, normalizeShearedChannels performs channel normalisation for each array.

References

G. K. Smyth and T. P. Speed. (2003) Normalization of cDNA microarray data. Methods 31:265-27

Examples

Run this code

#Read files into BeadSetIllumina-object
rPath <- system.file("extdata", package="beadarrayMSV")
BSDataRaw <- readBeadSummaryOutput(path=rPath,recursive=TRUE)

#Find indexes to sub-bead pools
beadInfo <- read.table(paste(rPath,'beadData.txt',sep='/'),sep='\t',
    header=TRUE,as.is=TRUE)
rownames(beadInfo) <- make.names(beadInfo$Name)
normInd <- getNormInd(beadInfo,featureNames(BSDataRaw))

#Pre-process
normOpts <- setNormOptions(minSize=10)
plotPreprocessing(BSDataRaw,normInd,normOpts,plotArray=1)
BSData <- preprocessBeadSet(BSDataRaw,normInd,normOpts)
pData(BSData)

Run the code above in your browser using DataLab