I'm working with a data set which consists of 3 patients. I have from each patients 9 samples and 3 samples were treated with treatment A, 3 samples with treatment B and 3 samples are the control.
I'm interested in the difference of Control vs. Treatment A and Treatment A vs Treatment B per patient and for all patients.
I used the following design for DESeq2:
coldata$combine <- factor(paste0(coldata$patient, "-", coldata$treatment)) dds <- DESeqDataSetFromMatrix(countData = Jcounts, colData = coldata, design = ~ combine)
I like the combination of the factors because it is easier use, you get your results with the results function and using list and recommended by Michael Love in a previous experiment
This design should be comparable with
dds <- DESeqDataSetFromMatrix(countData = Jcounts, colData = coldata, design = ~ patient + treatment + patient:treatment)
for my questions.
The PCA shows clear differences between the treatments with Control on left side, Treatment A in the middle and B on the right side of the x-axis (PC1) for all patients. Furthermore Patient 1 and 2 cluster together within different treatments. However, patient 3 is away from the other two patients on the y-axis (PC2) with 25 % variance.
With this picture in mind, can I use the following for the treatment comparison over all patients?
result_all <- results(dds, contrast = list(c("combineP1.A", "combineP2.A", "combineP3.A"), c("combineP1.Control", "combineP2.Control", "combineP3.Control")), listValues = c(1/3, -1/3))
I get about 6,000 genes with a padj of 0.1.
Or should I just do a design only with treatment
dds <- DESeqDataSetFromMatrix(countData = Jcounts, colData = coldata, design = ~ treatment)
and compare A vs Control. I get about 3,000 genes with a padj of 0.1 and ca. 98 % of the are in the 6,000 from above. This design doesn't use the patient difference. So I guess it is not the best solution.