Biostars beta testing.
Question: How to interprate edgeR result for different comparison
0
Entering edit mode

I want to find interaction between type (A and B) and dosage (6 and 26) in RNA-seq data. The design was:

(Intercept) typeA dosage6 typeA:dosage6
B26           1      0       0           0
B26           1      0       0           0
A26           1      1       0           0
A26           1      1       0           0
A26           1      1       0           0
A6            1      1       1           1
A6            1      1       1           1
A6            1      1       1           1
B6            1      0       1           0
B6            1      0       1           0
B6            1      0       1           0
B26           1      0       0           0

My differential expression analysis output result using edgeR R package is:

tag     typeA            dosage6       typeA.dosage6
ENSG00000000305        -1                  1                  -1
ENSG00000000308         1                  1                  -1
ENSG00000000340        -1                  1                  -1
ENSG00000000378         0                  0                   0
ENSG00000000420         0                  0                   0
ENSG00000000441         0                  0                   0
ENSG00000000486         1                 -1                  -1
ENSG00000001127        -1                 -1                   1

How I could understand if there is any interaction between type and dosage? in the other hand how this result could help me to find out the interaction?

Entering edit mode
0

russhh
♦ 4.4k
Entering edit mode
0
Contrasts
Levels        typeA  dosage6  typeA.dosage6
Intercept         0        0              0
typeWT            1        0              0
age6              0        1              0
typeWT.age6       0        0              1
abc
• 30
0
Entering edit mode

Could you try using the following design matrix and see if your results seem more reasonable, that is, if boxplots of genes with nonzero interaction term look like there is an interaction:

I find it more straightforward to encode binary variables (-1,1), with the interaction term as product. So I'd have:

(Intercept)  A-B    d6-d26    AB:d
B26           1      -1      -1          1
B26           1      -1      -1          1
A26           1      1       -1          -1
A26           1      1       -1          -1
A26           1      1       -1          -1
A6            1      1       1           1
A6            1      1       1           1
A6            1      1       1           1
B6            1      -1      1           -1
B6            1      -1      1          -1
B6            1      -1      1           -1
B26           1      -1      -1          1

With this design you can use your contrasts matrix without change (aside from relabelling the rows).