Question on DESeq2 nested model example
1
0
Entering edit mode
Guido Hooiveld ★ 4.0k
@guido-hooiveld-2020
Last seen 3 days ago
Wageningen University, Wageningen, the …

I am reading the DESeq2 manual (version of 16 Oct 2015), and have a question on the example used in section 3.12 (“Model matrix not full rank"), specifically on the nested model (3.12.1) that deals with testing group-specific effects of a treatment, while controlling for individual effects. The sample assignment of the nested model is this (copied from the manual to allow better reading of this post):

coldata

> coldata
grp ind cnd ind.n
1   X   1   A     1
2   X   1   B     1
3   X   2   A     2
4   X   2   B     2
5   Y   3   A     1
6   Y   3   B     1
7   Y   4   A     2
8   Y   4   B     2
>

The experimental design is:

design2 <- model.matrix(~ grp + grp:ind.n + grp:cnd)

resulting in this model matrix:

> design2

(Intercept) grpY grpX:ind.n2 grpY:ind.n2 grpX:cndB grpY:cndB

1           1    0           0           0         0         0

2           1    0           0           0         1         0

3           1    0           1           0         0         0

4           1    0           1           0         1         0

5           1    1           0           0         0         0

6           1    1           0           0         0         1

7           1    1           0           1         0         0

8           1    1           0           1         0         1

I understand that the coefficients “grpX:cndB” and “grpY:cndB” represent the effect of condition B vs A in group X resp. Y. Maybe a naive question, but is it is also possible to also extract the differences between the 2 groups? Thus what is the difference between groups X and Y at Condition A (and Condition B)? At first sight I thought this is represented by the 2nd coefficient (grpY), but then I realized this coefficient represents the average of condition A plus B in grpY. Is it thus not possible to make these between-group comparisons using this design?

Thanks,

Guido

deseq2 design matrix • 1.3k views
0
Entering edit mode

After giving my question some more thoughts, I think the difference between group X and Y at condition A is represented by the contrast:

grpY

and the difference between group X and Y at condition B:

(grpY:cndB - grpX:cndB + grpY)

... but please correct me if I am wrong.

<< EDIT >> I was wrong; see Michael's reply below.

0
Entering edit mode
@mikelove
Last seen 2 days ago
United States

You cannot directly compare X and Y because they do not share the same individuals. The design does not allow for that comparison. How do you know an X vs Y effect from a individual 1+2 vs 3+4 effect?

0
Entering edit mode

Well, I thought along the lines of an "unpaired" comparison between the individuals of grpX and grpY at condition A resp. B, whereas the effects of A and B are compared within both grps controlling for individual ("paired" analyses).

But as said before, I wasn't sure... Thanks!