Problem to understand Interactions at any time
Entering edit mode
vm.higareda ▴ 10
Last seen 3.0 years ago


I am having troubles to understand interactions in edgeR, probably this is simple but I can get it. I will be referring to 3.3.4 topic of edgeR user`s guide. The experimental design of this section is:


           Treat   Time 
Sample1  Placebo  0h
Sample2  Placebo  0h
Sample3  Placebo  1h
Sample4  Placebo  1h
Sample5  Placebo  2h
Sample6  Placebo  2h
Sample7   Drug    0h
Sample8   Drug    0h
Sample9   Drug    1h
Sample10  Drug    1h
Sample11  Drug    2h
Sample12  Drug    2h       

matrix is generated with

design <- model.matrix(~Treat * Time, data=targets)

             (Intercept) TreatDrug  Time1h  Time2h     TreatDrug:Time1h TreatDrug:Time2h
Sample1            1         0      0      0                0                0
Sample2            1         0      0      0                0                0
Sample3            1         0      1      0                0                0
Sample4            1         0      1      0                0                0
Sample5            1         0      0      1                0                0
Sample6            1         0      0      1                0                0
Sample7            1         1      0      0                0                0
Sample8            1         1      0      0                0                0
Sample9            1         1      1      0                1                0
Sample10           1         1      1      0                1                0
Sample11           1         1      0      1                0                1
Sample12           1         1      0      1                0                1


[1] "(Intercept)" "TreatDrug" "Time1h" "Time2h" [5] "TreatDrug:Time1h". "TreatDrug:Time2h"

My questions are:

What represent the intercept? is an average of expression of all samples including drug, placebo at all times?.

In the following example:

glmQLFTest(fit, coef=2)

According to matrix coef2= TreatDrug = Drug at time 0,1 and 2. Using coef=2 , Do I comparing TreatDrug vs Intercept ?

In biological sense what genes will be detecting doing this comparison?

I really appreciate your help.

RNA-seq edgeR interactions • 902 views
Entering edit mode
Last seen 4 hours ago
WEHI, Melbourne, Australia

The intercept term is not explained in Section 3.3.4 of the edgeR User's Guide because you should not be conducting any tests or contrasts using that term. It represents the average log2 CPM of the placebo treatment at time 0h. With the default the parametrization in R, the intercept always represent the average with all factors at their baseline levels.

coef=2 is as explained in the User's Guide. It will detect genes that are DE between drug and placebo at 0 hours.

Entering edit mode

In that sense coef= 3 are the effects of placebo at 0 hours vs. placebo at 1 hour? coef= 4, are the effects of placebo at 0 hours vs. placebo at 2 hours? and "TreatDrug:Time1h" = coef=5 will be placebo 1 hour vs. Drug 1 hour.

I have been reading the User's Guide, it is well documented but I am slow learner,


Entering edit mode

Section 3.3 of the User's Guide presents three design matrices to analyse this experimental design. We recommend that you define each treatment combination as a group (Section 3.3.1) because it is by far the most transparent and easier to interpret, especially if you are new to statistical formula.

We cover the interaction formula in Section 3.3.4 only because this is the traditional statistical approach. However the interaction formula is really only useful for testing interactions, represented by coefficients 5:6. The other coefficients are not any scientific use and it is probably not worth your while trying to understand what they mean. In practtice, you should not be conducting tests for any of the coefficients 1:4 from the interation formula. You can get any results you need in a clearer way from the group means formula in Section 3.3.1.


Login before adding your answer.

Traffic: 667 users visited in the last hour
Help About
Access RSS

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6