Dear Michael

This is a kind of a follow up of a previous post (https://support.bioconductor.org/p/59899/).

I am trying to use DESeq2 for counts data transformation. My data is coming from mapping to a de-novo assembled transcriptome so we therefore use eXpress for abundance estimation. When I try to run varianceStabilizingTransformation() or DESeq()received the error:

“Error in estimateSizeFactorsForMatrix(counts(object), locfunc, geoMeans = geoMeans) : every gene contains at least one zero, cannot compute log geometric means”

I can understand that a log transformation would be problematic if there are zeros but it makes sense that some genes, in some conditions, have zero counts estimated (i.e. are not expressed). Am I missing something or is there a workaround like the y=log2(y+ 1)?

Sincerely,

Moran   

R code

data <- read.table("matrix", header = T, row.names = 1, com= '')

data <- round(data)

data <- data[rowSums(data) >= 13,]

conditions = data.frame(conditions=factor(c(rep("inter ", 5),

                                                                             rep("late ", 6),

                                                                             rep("early ", 10),

                                                                             rep("flo", 4),

                                                                             rep("se", 4), )))

rownames(conditions) <- colnames(data)

ddsTable <- DESeqDataSetFromMatrix(countData = data,

                                   colData = conditions,

                                   design = ~ conditions)

vsd <- varianceStabilizingTransformation(ddsTable)

> as.data.frame(colData(ddsTable))

                                                conditions

X10DP            inter

X10DP           inter

X10DP            inter

X10DP            inter

X10DP            inter

X21DP             late

X21DP             late

X21DP             late

X21DP             late

X21DP             late

X21DP             late

X4DP             early

X4DP             early

X4DP             early

X4DP             early

X4DP             early

X4DP             early

X4DP             early

flo_1           flo

flo_2           flo

flo_3           flo

flo_4           flo

se_1               se

se_2               se

se_3               se

se_4               se

> sessionInfo()

R version 3.1.1 (2014-07-10)

Platform: x86_64-apple-darwin13.1.0 (64-bit)

locale:

[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

attached base packages:

[1] parallel  stats     graphics  grDevices utils     datasets  methods   base    

other attached packages:

 [1] BiocParallel_0.6.1        BiocInstaller_1.14.3      DESeq2_1.4.5            

 [4] RcppArmadillo_0.4.450.1.0 Rcpp_0.11.3               GenomicRanges_1.16.4     

 [7] GenomeInfoDb_1.0.2        IRanges_1.22.10           BiocGenerics_0.10.0     

[10] ggplot2_1.0.0             edgeR_3.6.8               limma_3.20.9            

loaded via a namespace (and not attached):

 [1] annotate_1.42.1      AnnotationDbi_1.26.1 base64enc_0.1-2      BatchJobs_1.4      

 [5] BBmisc_1.7           Biobase_2.24.0       brew_1.0-6           checkmate_1.5.0    

 [9] codetools_0.2-9      colorspace_1.2-4     DBI_0.3.1            digest_0.6.4       

[13] evaluate_0.5.5       fail_1.2             foreach_1.4.2        formatR_1.0        

[17] genefilter_1.46.1    geneplotter_1.42.0   grid_3.1.1           gtable_0.1.2       

[21] iterators_1.0.7      knitr_1.7            lattice_0.20-29      locfit_1.5-9.1     

[25] MASS_7.3-35          munsell_0.4.2        plyr_1.8.1           proto_0.3-10       

[29] RColorBrewer_1.0-5   reshape2_1.4         RSQLite_0.11.4       scales_0.2.4       

[33] sendmailR_1.2-1      splines_3.1.1        stats4_3.1.1         stringr_0.6.2      

[37] survival_2.37-7      tools_3.1.1          XML_3.98-1.1         xtable_1.7-4       

[41] XVector_0.4.0

hi Moran,

The median ratio method defined in the first DESeq paper requires calculation of the geometric mean of a row, which is 0 if a single count has a 0, and then taking the ratio of a sample to the geometric mean.

I would first filter the dataset to remove rows with very few counts (with x set to 5 say):

dds <- dds[ rowSums(counts(dds)) > x, ]

Then one alternative estimator I have used when there are no rows without a zero is:

cts <- counts(dds)
geoMeans <- apply(cts, 1, function(row) if (all(row == 0)) 0 else exp(mean(log(row[row != 0]))))
dds <- estimateSizeFactors(dds, geoMeans=geoMeans)

Hi Michael

Thank you very much for the reply, and fix. I do not have access to my data now but will try give a try next week when I'm back

Moran