DESeq2 - using contrasts vs. multiple runs for DE analysis

[rutgers2015]

Hello all,

I have a time series dataset with 4 time points (no other factors besides time). What is the better, more statistically sound strategy to test genes in each time point against one another for DE:

1) Run DESeq on the entire data set (4 time points * 3 replicates each) and extract each contrast using results(dds, contrast = c("time",x,y) for a total of 6 results tables ( 1 v 2, 1 v 3, 1 v 4, 2 v 3, 2 v 4, 3 v 4)

2) Run DESeq six times on six different data sets, with 2 time points in each, and use the results table from each run.

I plan on extracting all genes that have at least LFC of 1 and padj < 0.05 . Applying this filter on a result table from (2) gives me less genes than on the corresponding table in (1). Does this mean that running the analysis individually for each comparison is more stringent?

Any help would be appreciated.

[dpryan]

Option 1 is generally the better idea. Option 2 isn't more stringent, you just have fewer total samples so that'll affect your variance estimates and degrees of freedom.

[friducha]

Hi,

I am in a similar situation with my data.

I have three conditions, with 4 replicates for CondA, 2 replicates for CondB, and 2 replicates for CondC.

When I run DESeq using all 8 samples, I get 197 DE genes with a padj < 0.1 for contrast CondB/CondA.

However, when I take out the samples from CondC from the counts matrix, for the same contrast CondB/CondA, I get only 170 DE genes.

So in my case, the second method is more stringent... although I would like to understand why this happens.

Thanks for your help.