High log2foldchange in Deseq2

Question

[amenshamim]

I have ICGC raw read counts. I applied DeSeq2 for expression analysis. I have two conditions, one is normal sample and other one is tumor sample for each gene.

conds <- as.factor(c('Normal','Tumor'))
coldata <- data.frame(row.names=colnames(q),conds)
dds=DESeqDataSetFromMatrix(countData = q, colData = coldata, design = ~conds )
dds <-estimateSizeFactors(dds,type="poscounts")
dds <- DESeq(dds)
res <- results(dds)

The results show that log2foldchange is maximum 30 and minimum -30. My question is, foldchange is very high in my study. Most papers showed 1 to 4 log2foldchange. Anyone can figure out why fold change is that much high. The highest and lowest values are showing 30 or -30 log2foldchange, respectively. Is there any specific reason to give a cutoff of 30?

Answer 1

[Kevin Blighe]

Hi,

It's implied by your code that you are comparing 1 tumour versus 1 normal sample. That's virtually nil statistical power.

Also, why are you using type = 'poscounts' for estimateSizeFactors()? - 'ratio' is the default.

You are additionally missing the fold-change shrinkage step - see: Quick start.

Also, a tip: I would not run results() blind in the way that you are doing. Always select a coeficient via the name parameter (or a contrast via Contrasts) so that you are absolutely sure which results are being calculated and returned. again, please refer to the Quick Start and, indeed, the remainder of the DESeq2 vignette.

Kevin

[amenshamim]

Thank you for your response. In my dataset, Tumor and normal datset are not paired. For example, 4 tumors vs 2 normal. Some normal dataset contains zero value. That's why I used poscounts to deal with it. What would be your suggestion, In this condition, I have to use default 'ratio' or poscounts?

[Kevin Blighe]

4 tumour and 2 normal? - but this line indicates that you just have 2 samples:

conds <- as.factor(c('Normal','Tumor'))

You may want to first verify your code.

I see no major reason for using poscounts - please just try the default. Genes with 0 values across all samples should be removed prior to running DESeq2