vm.analysis {varmixt}R Documentation

Variance mixture analysis on unpaired data

Description

Performs variance mixture analysis on unpaired data

Usage

vm.analysis(geneId,cont,treat,filename=NULL,gene.anot=NULL,badqual=NULL,qualtol=NULL,n.mixt=NULL,
                      center=TRUE,loess.cor=FALSE,min.rep=2,penalty=c("AIC","BIC")[2],display=TRUE,stop.crit.1=1.e-6,
                      criterion.1=c("likelihood","parameter")[1],criterion.2=c("likelihood","parameter")[1],
                      stop.crit.2=1.e-8)

Arguments

geneId vector with the identifier of each gene (one gene per row)
cont matrix of gene log-intensity in condition 1. The matrix has one gene per row and one replicate per column
treat matrix of gene log-intensity in condition 1
filename filename for export. If NULL no files are exported
gene.anot data.frame with further gene anotations
badqual integer matrix of number of bad quality observation per gene and replicate. Do not use badqual with different number of replicates in each condition
qualtol integer. Genes with more than qualtol bad quality observations are removed from the analysis. This argument is ignored is Badqual is NULL
n.mixt integer. Number of component in the mixture model. If n.mixt=NULL the optimal number of components is computed by the function
center Logical. If True each array*condition is centered to have a 0 mean log-intensity.
loess.cor Logical. if True then loess transformation of the data is performed. If False no loess transformation is performed
min.rep Minimum number of non-missing value per gene and per condition. Must be at least 2.
penalty Character. Either "BIC" or "AIC". The criterion for choosing the number of variance groups.
display Boolean. Should the result of model fitting be displayed on the screen, on the fly.
stop.crit.1 The stopping relative precision limit for stopping EM algorithm
criterion.1 either "likelihood" or "parameter". The criterion for stopping EM algorithm while trying to determine the number of variance groups
criterion.2 Either "likelihood" or "parameter". The criterion for stopping EM algorithm while trying to estimate the parameters.
stop.crit.2 The stopping relative precision limit for stopping EM algorithm

Details

We highly recommend the use of pre-normalized data. The function only centers the data set. It can only perform very simple global array by array lowess transform. Typically the badqual matrix is generated by couting the number of times of gene is under the detection threshod (background) or at the saturation level on each physical array. A gene with more than "qualtol" bad quality observation is removed from the analysis. The corrected p-value (controlling FWER) are computed using the Bonferroni correction.\ The EM algorithm stops when the relative increase in likelihood or relative maximum absolute difference in parameter value becomes less than the stop.crit value. If t is the iteration number, and the criterion is likelihood then the stop criterion is (log-like[t]-log-like[t-1])/log-lik[t-1]<stop.crit

Value

A gene expression data object with the results of the variance mixture analysis

geneid the vector of gene names
raw.cond1 a matrix of the raw log-intensity in condition 1
raw.cond2 a matrix of the raw log-intensity in condition 2
cond1 a matrix of the normalized log-intensity in condition 1
cond2 a matrix of the normalized log-intensity in condition 2
stat1 a data.frame of results of homoscedastic and gene-specific model
stat1.call call that generated the stat1 data.frame
df number of degrees of freedom
var.to.vardelta factor for computing the test statistic variance given the gene variance
residual matrix of the residual
stat2 a data.frame of results of the variance mixture model analysis
stat2.call call that generated the stat2 data.frame
param a data.frame of value of the variance mixture paramters
call The call to the function
ppost The matrix of posterior probability that a gene belongs to each variance component. One row per gene, one column per variance component.
choose.nmixt A data frame with some information of models with growing number of components

Author(s)

Paul Delmar and Julie Aubert

References

P. Delmar, Robin, S., Tronik-Le Roux S. and Daudin J.-J. (2005) Mixture model on the variance for the differential analysis of gene expression data, JRSS series C, 54(1), 31:50

See Also

fdr.an, export.res, plotvm, qqplot.vm, plotrm, boxplotvm , compute.dif

Examples

data(apo.data.vm)
res.apo<-vm.analysis(geneId=apo.data.vm$apo.geneid,cont=apo.data.vm$apo.cond1,
                     treat=apo.data.vm$apo.cond2)
fdr.an(res.apo,0.01)
res.apo.data.frame<-export.res(res.apo)
par(mfrow=c(2,2))
plotvm(res.apo)
plotrm(res.apo)
boxplotvm(res.apo)
qqplot.vm(res.apo)

[Package varmixt version 0.2-4 Index]