Deseq Handle Samples Without Replicates

Question

[camelbbs]

I've tried the package DESeq and want to apply in our datasets: two samples without replicates.

I first summarized the counts in a table: brain liver uc001aaa.3 0 0 uc001aac.4 0 0 uc001aae.4 0 0 ... ...

By following code:

over<-read.table("wokao.txt",header=T,row.names=1)
conds<-c("brain","liver")
cds<-newCountDataSet(over,conds)
cds<-estimateSizeFactors(cds)
cds<-estimateDispersions(cds,method="blind",sharingMode="fit-only")

Error in parametricDispersionFit(means, disps) :

Parametric dispersion fit failed. Try a local fit and/or a pooled estimation. (See '?estimateDispersions')

cds2=cds[,c("brain","liver")]
cds2=estimateSizeFactors(cds2)
cds2=estimateDispersions(cds2,method="blind",sharingMode="fit-only")

Error in parametricDispersionFit(means, disps) :

Parametric dispersion fit failed. Try a local fit and/or a pooled estimation. (See '?estimateDispersions')

I have no idea why there is the error. May you give some guides.

sessionInfo()
R version 2.15.1 (2012-06-22)
Platform: x86_64-pc-mingw32/x64 (64-bit)


locale:
[1] LCCOLLATE=EnglishUnited States.1252  LCCTYPE=EnglishUnited States.1252    LCMONETARY=EnglishUnited States.1252
[4] LCNUMERIC=C                           LCTIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] DESeq1.8.3        locfit1.5-8       Biobase2.16.0     BiocGenerics0.2.0

loaded via a namespace (and not attached):
 [1] annotate1.34.1      AnnotationDbi1.18.4 DBI0.2-5            genefilter1.38.0    geneplotter1.34.0 

 [6] grid2.15.1          IRanges1.14.4       lattice0.20-10      RColorBrewer1.0-5   RSQLite0.11.2 

[11] splines2.15.1       stats42.15.1        survival2.36-14     XML3.95-0.1         xtable_1.7-0 

[Steve Lianoglou]

Please update your question by pasting in the output of sessionInfo() after the DESeq package has been loaded. It is also suggesting that you try a local fit for your estimation, so why not try that? Look in ?estimateDispersions to figure out how.

[camelbbs]

Thanks, But I don't know how to use local fit in this code

[camelbbs]

While I add cds<-estimateDispersions(cds,method="blind",sharingMode="fit-only",fitType="local") It works and can be used in res<-nbinomTest(cds,"brain","liver")

Is that right?

[Steve Lianoglou]

For what definition of "right"? I mean, it seems to "work", so ... what else can I tell you? Without any replicates, it's hard to say that something is right.

I'm curious why the other approach fails, though. You might want to upgrade to the latest release of bioconductor (and its related packages). The current release of DESeq is 1.10.1. The very top of the Bioconductor Install Page shows how to upgrade to Bioconductor 2.11:

source("http://bioconductor.org/biocLite.R")
biocLite("BiocUpgrade")

[camelbbs]

Thanks a lot. ~~

[Neilfws]

Hint: when you begin lines in your question with ">", it's formatted as a blockquote; probably not what you want. Instead, prefix lines of code with 4 spaces. Fixed it for you.

Answer 1

[earonesty]

edgeR handles samples without replicates well, but the question is why? Unless you are looking at platform technical variation, you won't find any biologically meaningful results.