Design Matrix for multiple variables in Edge R
0
1
Entering edit mode
@6fc08061
Last seen 3.4 years ago

Hi ,

I am working on longitudinal analysis. I have to perform DE between 2 groups of patients between T0-->T1

The variables in my design are Time (Visit1==T0, Visit2==T1), Sex(Male, Female), Status(RA, C) and Subject(Patient Numbers).

Samples are paired and each group has male and female samples This is how I set up my design matrix.


LTR_ID  Visit Disease Sex  Subject
LTR010  1       0       F   1
LTR093  2       0       F   1
LTR012  1       1       M   1
LTR094  2       1       M   1
LTR017  1       1       M   2
LTR095  2       1       M   2
LTR096  1       1       F   3
LTR097  2       1       F   3

design <- model.matrix(~status + sex + status:subject + status:time)
colnames(design)

[1] "(Intercept)"        "statusRA"           "sexM"               "statusC:subject2"
[5] "statusRA:subject2"  "statusC:subject3"   "statusRA:subject3"  "statusC:subject4"
[9] "statusRA:subject4"  "statusC:subject5"   "statusRA:subject5"  "statusC:subject6"
[13] "statusRA:subject6"  "statusC:subject7"   "statusRA:subject7"  "statusC:subject8"
[17] "statusRA:subject8"  "statusC:subject9"   "statusRA:subject9"  "statusC:subject10"
[21] "statusRA:subject10" "statusRA:subject11" "statusRA:subject12" "statusRA:subject13"
[25] "statusRA:subject14" "statusRA:subject15" "statusRA:subject16" "statusRA:subject17"
[29] "statusRA:subject18" "statusRA:subject19" "statusRA:subject20" "statusRA:subject21"
[33] "statusRA:subject22" "statusRA:subject23" "statusC:time2"      "statusRA:time2"

#testing for DE with patients at T0 vs T1

y  <- estimateDisp(y, design)

==>Error in glmFit.default(sely, design, offset = seloffset, dispersion = 0.05, :
Design matrix not of full rank. The following coefficients not estimable: statusRA:subject19


I get the above error. I am doing a pair wise analysis where I want to check for DE between patients at T0 vs T1. I need to account for Sex as a covariate and do the analysis. Looks like I am setting up the design matrix wrong. I have 2 questions, Does doing pairwise analysis account for gender effect? If not, What would be the best way to set ithe design?