Question

Getting a large intercept with DESeq2

0

Entering edit mode

Denise • 0

@860ad694

Last seen 3 hours ago

Spain

Hi,

I have an RNA-seq dataset where a treatment has been applied and treatment effects measured in replicate across several time points post-exposure. I have 3 treatment and 3 control replicates for each time point. I then run DESeq2's LR test below to find genes affected by treatment at any time point.

full_model ~ extraction_batch + SVs + timepoint + treatment + treatment:timepoint 
reduced_model ~ extraction_batch + SVs + timepoint

However, when I plot the model coefficients obtained using the coeff() function, I notice that the intercept is huge across all genes, whereas the other coefficients / LFCs are much smaller by comparison. What does this mean? Am I doing something wrong? Given the biological question I want to answer, I am positive these are the models I should use.

Thank you in advance for your advice.

Denise

sessionInfo( )

```

DESeq2 DifferentialExpression TimeCourse • 99 views

ADD COMMENT • link 3 days ago • updated 4 hours ago Denise • 0

score 0 · Answer 1 · 2025-11-14

0

Entering edit mode

James W. MacDonald 68k

@james-w-macdonald-5106

Last seen 4 hours ago

United States

The default parameterization in R is a treatment-contrasts parameterization that sets one of your groups as the baseline. So the intercept is the average expression value (I believe logCPM for DESeq2, but don't quote me on that) for the baseline group, which you can identify by inspecting your design matrix.

ADD COMMENT • link 3 days ago James W. MacDonald 68k

0

Entering edit mode

As a trivial example, consider this:

> model.matrix(~Treatment, data.frame(Treatment = factor(rep(1:3, each = 4))))
   (Intercept) Treatment2 Treatment3
1            1          0          0
2            1          0          0
3            1          0          0
4            1          0          0
5            1          1          0
6            1          1          0
7            1          1          0
8            1          1          0
9            1          0          1
10           1          0          1
11           1          0          1
12           1          0          1

The baseline in this case is treatment 1.

ADD REPLY • link 3 days ago James W. MacDonald 68k

0

Entering edit mode

Thanks James -- Denise you can check the "note on factor levels" in the vignette.

And I'd recommend plotCounts() for looking at individual genes.

ADD REPLY • link 3 days ago Michael Love 43k

0

Entering edit mode

Hi,

Thank you very much, James and Michael, for getting back to me! I noticed the large intercept when plotting the coefficients returned by coef() with pheatmap(), as used in the DESeq2 vignette.

I know how to interpret each coefficient returned by resultsNames() after running LRT, but not the intercept it seems! So, in my case, the intercept is the average gene expression in controls at time point A!

Thank you.

ADD REPLY • link 4 hours ago Denise • 0