Entering edit mode
Jaaved Mohammed
▴
30
@jaaved-mohammed-6011
Last seen 9.6 years ago
I have another related question regarding contrast design in this
"tissue-specific expression" framework. I get different results if I
merge
several tissues into one group in the contrast matrix instead of
assigning
each tissue a unique group. I think redesigning the contrast like this
simplifies the model and hence explains the difference in results.
Here is the new contrast design:
z = as.character(y$samples$group)
z[which(z != "tissue1")] = "group2" # <-- label everything not
tissue1 as
group2
z[which(z == "tissue1")] = "group1" # <-- keep tissue1 as a separate
label, group1
#Design the new contrast matrix to find diff exp between group1 and
group2
fac = factor(z)
design = model.matrix(~0+fac)
colnames(design) = levels(fac)
cont.matrix = makeContrasts(diff=group1 - group2, levels = design)
I cannot determine if the results are better or worse by looking at
the top
genes. Should I really be consolidating disparate tissues like this,
or is
the model richer by keeping all tissues separate?
Thanks,
Jaaved
[[alternative HTML version deleted]]