DESeq2: overlap differentially expressed genes with and without interaction term
1
0
Entering edit mode
xuscha • 0
@d1c27c62
Last seen 4 weeks ago
Austria

Hello everyone!

I have a data set with 3 treatment types+a control sample(DMSO), each at 3 incubation time points (24, 48, 72hr). I am having trouble understanding the large discrepancy when extracting the results for the treatment group Ibr using the two designs below.

Only 36% of the resulting DEGs are shared between the two lists, is this expected?

I am using DESeq2_1.30.1

Any help would be greatly appreciated, thanks!

design<-~passage+inc_time+treatment
dds <- DESeqDataSetFromMatrix(countData=countData, colData=colData, design=design)
dds$inc_time <- relevel(dds$inc_time, ref = "24")
dds <- DESeq(dds)

design<-~passage+inc_time+treatment+inc_time:treatment
dds1 <- DESeqDataSetFromMatrix(countData=countData, colData=colData, design=design)
dds1$inc_time <- relevel(dds1$inc_time, ref = "24")
dds1 <- DESeq(dds1)

venn.diagram(x = list(rownames(R1),rownames(R2)),
category.names= c("Full~passage+inc_time+treatment","FULL~passage+inc_time+treatment+inc_time:treatment"),
width = 800,height = 600 ,
fill = myCol[c(1:2)],lty = 'blank', imagetype="png",
fontface = "bold",cat.default.pos = "outer",cat.pos = c(-15,15),
filename="deseq_design_comparison.png",cex = .5,cat.cex = .2)


DESeq2 • 211 views
0
Entering edit mode
swbarnes2 ★ 1.2k
@swbarnes2-14086
Last seen 1 day ago
San Diego

You are asking the data two different questions, of two different sets of samples, so I'm not sure why you would expect the same answer for both questions.

0
Entering edit mode

Not the same but I was expecting the overlap to be much higher. I am unsure what you mean with "two sets of samples"? I am guessing you mean the difference in model matrix?

I want to find genes DE in first and foremost due to treatment. Then I want to further test if these are influenced by the factor time.

I am uncertain if I am interpreting the following two comparisons correctly as both study the effect of the factor "time":

1) Find genes not influenced by the factor time (H0: Excluding the factor "time" results in as good of a fit):

~passage+inc_time+treatment VS ~passage+treatment


Extracting DE genes that also coincide with low LRT p-values will result in genes significantly effected by treatment AND by time?

2) Find genes not influenced by the interaction between time and treatment (even though time on its own is still a factor in the model): (H0: "how time is interacting with treatment" has no influence on the goodness of my fit)

~passage+inc_time+treatment+inc_time:treatment VS ~passage+inc_time+treatment


Extracting DE genes that also coincide with low LRT p-values will result in genes significantly effected by treatment, and this particular effect of treatment is furthermore changing over time?