DESeq2 - controlling for other variables
1
0
Entering edit mode
@yonatan-amzaleg-19022
Last seen 17 months ago
Los Angeles, CA, University of Southern…

Hello Everybody,

I have been performing RNA-seq analyses using DESeq2 a particular way and I just wanted to verify whether my logic is sound.

Basically my experimental condition is such - I have a cell line in which there is a heterozygous mutation at a gene of interest. This Heterozygous mutation (amongst other things) leads to constitutive activation version of the gene of interest. I wanted to do an RNA-seq analyses to compare the transcriptome of this cell with the activating mutation to the same WT cell line, but activated by a ligand.

Because the mutant cell line is heterozygous it still has one copy that is WT and my concern is amongst the differentially expressed genes are genes in which are upregulated by WT in the vehicle setting compared to WT with the ligand and not Mutant vs WTligand. To combat this we also performed RNA-seq on WTveh and then performed a DE analysis with WTligand. I decided (maybe incorrectly) that if it has a padj < 0.05 and log2foldchange > 1 it is considered upregulated in unliganded vs liganded in WT cells. This was the list I removed from the Mutant vs WTligand comparison after I performed the DE analysis there.

Is this an appropriate approach? Does anyone have a better suggestion? I actually did this exact thing to two different cell lines in which there is a corresponding heterozygous mutation (separately, of course)

I appreciate everyone's help,

Yonatan

deseq2 DEG covariate study design variables • 353 views
1
Entering edit mode
@mikelove
Last seen 12 hours ago
United States

To generalize, are you saying you are interested in genes for which the LFC for B vs A is significant, while C vs A is small?

Instead of padj > .05, try altHypothesis="lessAbs" instead. See another recent Bios post (yesterday).

0
Entering edit mode

Thank you so much, Michael!

I'm curious, because the way I've processed this experiment in r is the following

dds <- DESeq(dds)

then I check condition upregulated in "A" vs "B" (My primary analysis)

dds <- results(dds contrast=c("Condition", "A", "B"))
dds <- dds[ which(dds$padj < .05), ] dds <- as.data.frame(dds [ order(dds$padj), ])
dds <- subset(dds, log2FoldChange > 1)


I want to make sure that the upregulated genes in A over B, aren't upregulated in C compared to B so I do the same as above looking at upregulatedin C compared to B, then remove that from my final list. What you are saying is in the C vs B analysis, don't look at padj < 0.05 but rather altHypothesis="lessAbs" then remove these genes from the list I made above?

Sorry for the confusion,

Yonatan

0
Entering edit mode

You can either:

Remove genes from list in which LFC was large (e.g. remove genes when you had rejection of null of LFC=0), or you can actually test for LFC being small in the C vs A comparison (but then you have to specify what "small" means).

The approach is up to you. Note that, in the first approach, failure to reject the null does not mean the LFC is small, as it could also depend on power.