Question: Clarification on contrast levels in DESeq2
0
gravatar for Menachem Sklarz
18 months ago by
European Union
Menachem Sklarz10 wrote:

Hi,

I was running DESeq2 with the simplest possible design: one factor (called group) with 5 levels: A-E, where A is the reference level.

I run the model thus:

dds <- DESeqDataSetFromMatrix(countData = counts_data,
                              colData = col_data,
                              rowData = row_data,
                              design = ~ group)
dds <- DESeq(dds)

To test pairwise contrasts, I do the following:

b<-results(dds,contrast = c("group","A","B"))

etc.

However, I noticed several strange things:

  1. For the A-B comparison, I get significant genes when all samples in groups A and B have zero counts! It looks as if the comparison is against the Intercept, rather than level A, even though the results object says the data is for A vs. B comparison.
  2. When I change the reference level to E, I do not get these significant genes.
  3. When testing C vs. D, I get different results depending on whether A or E are the refenece level.
  4. I get a significant enrichment (adjusted p-value of 1.06E-04) when the reads in group A are 0 (for two samples) and the reads in group B are c(1,2,2). Does this make sense? Aren’t extremely low-count genes supposed to be filtered out in the “independent filtering” stage?

Note: I solved issues 1-3 by setting betaPrior to TRUE and modelMatrixType to "expanded".

Thanks a lot! Menachem

rna-seq deseq2 • 845 views
ADD COMMENTlink modified 17 months ago by Biostar ♦♦ 20 • written 18 months ago by Menachem Sklarz10

An obvious comment: why are you even using data for samples that are all zero?

ADD REPLYlink modified 17 months ago • written 17 months ago by Kevin Blighe52k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1982 users visited in the last hour