evalScoring: Score differential expression, assess significance, and smooth scores along the chromosome

Description

This function computes for all genes on one chromosome the regularized t-statistic to score differential gene expression for two given groups of samples. Additionally these scores are computed for a number of permutations to assess significance. Afterwards these scores are smoothed with a given kernel along the chromosome to give scores for chromosomal regions.

Usage

evalScoring(data, class, chromosome, nperms=1000, permute="labels", pcompute="empirical", subset=NULL, newlabels=NULL,kernel=rbf,kernelparams=NULL,cross.validate=TRUE, paramMultipliers=2^(-4:4),ncross=10,step.width=100000, memory.limit=TRUE, verbose=TRUE)

Arguments

data

Gene expression data in the MACAT list format. See data(stjude) for an example.

class

Which of the given class labels is to be analyzed

chromosome

Chromosome to be analyzed

nperms

Number of permutations

permute

Method to do permutations. Default 'labels' does permutations of the class labels, which is the common and faster way to assess significance of differential expression. The altenative 'locations' does permutations of gene locations, is much slower and right now should be considered preliminary at best.

pcompute

Method to determine the p-value for differential expression of each gene. Is only evaluated if the argument permute='labels' and in that case passed on to the function scoring

subset

If a subset of samples is to be used, give vector of column- indices of these samples in the original matrix here.

newlabels

If other labels than the ones in the MACAT-list-structure are to be used, give them as character vector/factor here. Make sure argument 'class' is one of them.

kernel

Choose kernel to smooth scores along the chromose. Available are 'kNN' for k-Nearest-Neighbors, 'rbf' for radial-basis-function (Gaussian), 'basePairDistance' for a kernel, which averages over all genes within a given range of base pairs around a position.

kernelparams

Additional parameters for the kernel as list, e.g., kernelparams=list(k=5) for taking the 5 nearest neighbours in the kNN-kernel. If NULL some defaults are set within the function.

cross.validate

Logical. Should the paramter settings for the kernel function be optimized by a cross-validation?

paramMultipliers

Numeric vector. If you do cross-validation of the kernel parameters, specify the multipliers of the given (standard) parameters to search over for the optimal one.

ncross

Integer. If you do cross-validation, specify how many folds.

step.width

Defines the resolution of smoothed scores on the chromosome, is in fact the distance in base pairs between 2 positions, for which smoothed scores are to be calculated.

memory.limit

If you have a computer with lots of RAM, setting this to FALSE will increase speed of computations.

verbose

logical; should function's progress be reported to STDOUT ?; default: TRUE.

Value

original.geneid: Gene IDs of the genes on the chosen chromosome, sorted according to their position on the chromosome
original.loc: Location of genes on chromosome in base pairs from 5'end
original.score: Regularized t-score of genes on chromosome
original.pvalue: Empirical p-value of genes on chromosome. How often was a higher score observed than this one with random permutations? In other words, how significant seems this score to be?
steps: Positions on the chromosome in bp from 5', for which smoothed scores have been computed.
sliding.value: Smoothed regularized t-scores at step-positions.
lower.permuted.border: Smoothed scores from permutations, lower significance border, currently 2.5%-quantile of permutation scores.
upper.permuted.border: Smoothed scores from permutations, upper significance border, currently 97.5%-quantile of permutation scores.
chromosome: Chromosome, which has been analyzed
class: Class, which has been analyzed
chip: Identifier for used microarray

Details

Please see the package vignette for more details on this function.

Examples

Run this code

    data(stjd) # load example data

    # if you have the data package 'stjudem' installed,
    #  you should work on the full data therein, of which
    #  the provided example data, is just a piece
    #loaddatapkg("stjudem")
    #data(stjude)

    # T-lymphocyte versus B-lymphocyte on chromosome 1, 
    #  smoothed with k-Nearest-Neighbours kernel(k=15), 
    #  few permutations for higher speed
    chrom1Tknn <- evalScoring(stjd,"T",chromosome="1",permute="labels",
    nperms=100,kernel=kNN,kernelparams=list(k=15),step.width=100000)

    # plotting on x11:
    if (interactive())
       plot(chrom1Tknn)

    # plotting on HTML:
    if (interactive())
       plot(chrom1Tknn,"html")

Run the code above in your browser using DataLab