Intensity and spatial normalization using robust neural networks fitting

Description

This function normalizes a batch of cDNA arrays by removing the intensity and spatial dependent bias.

Usage

1
maNormNN(mbatch,w=NULL,binWidth=3,binHeight=3,model.nonlins=3,iterations=100,nFolds=10,maplots=FALSE,verbose=FALSE) 

Arguments

mbatch

A marrayRaw or marrayNorm batch of arrays.

w

Weights to be assigned to each spot. If provided, it should be a vector with the same length as maNspots(mbatch).

binWidth

Width of the bins in the X direction (spot column) in which the print tip will be divided in order to account for spatial variation. Max value is maNsc(mbatch), Min value is 1. However if it is set to a number larger than maNsc(mbatch)/2 (so less than two bins in X direction) the variable X will not be used as predictor to estimate the bias.

binHeight

Height of the bins in the Y direction (spot row)in which the print tip will be divided in order to account for spatial variation. Max value is maNsr(mbatch), Min value is 1. However if it is set to a number larger than maNsr(mbatch)/2 (so less than two bins in Y direction) the variable Y will not be used as predictor to estimate the bias.

model.nonlins

Number of nodes in the hidden layer of the neural network model.

iterations

The number of iterations at which (if not converged) the training of the neural net will be stopped.

nFolds

Number of cross-validation folds. It represents the number of equal parts in which the data from a print tip is divided into: the model is trained on nFolds-1 parts and the bias is estimated for one part at the time. Higher values improve the results but increase the computation time. Ideal values are between 5 and 10.

maplots

If set to "TRUE" will produce a M-A plot for each slide before and after normalization.

verbose

If set to "TRUE" will show the output of the nnet function which is training the neural network models.

Details

This function uses neural networks to model the bias in cDNA data sets.

Value

A marrayNorm object containing the normalized log ratios. See marrayNorm class for details

Author(s)

Tarca, A.L.

References

A. L. Tarca, J. E. K. Cooke, and J. Mackay. Robust neural networks approach for spatial and intensity dependent normalization of cDNA data. Bioinformatics. 2004,submitted.

See Also

compNorm,nnet

Examples

1
2
3
4
5
6
7
# Normalization of swirl data
data(swirl)
# print-tip, intensity and spatial normalization of the first slide in swirl data set
swirlNN<-maNormNN(swirl[,1])   

#do not consider spatial variations, and display M-A plots before and after normalization
swirlNN<-maNormNN(swirl[,1],binWidth=maNsc(swirl),binHeight=maNsr(swirl),maplots=TRUE)