Entering edit mode
Hello! I am looking for DE genes comparing condition1 vs control. I have 3 biological replicates per each case, so 6 in total. I am using this design:
design=~ sex + condition
and I am filtering low expressed genes in this way:
smallestGroupSize <- 3
keep <- rowSums(counts(dds0) >= 10) >= smallestGroupSize
Then I get my DE that looks like this, with lfc=+/-1.5 and padj<0.005
But when i then look at raw counts and tpm of Slfn5 looks like this:
Why do I get this high padjusted value for this gene?
Hello Michael, thanks for your answer. Yes, I have only 3 samples :(. Dropping sex would remove those genes but at the same time does not include other DE genes. For example: is not included. This is also valid the other way around with
design=~genotype
. I was wondering if could be a possibility to use a more stringent filtering based on condition and variance for the multivariate model, what do you think?TPM is not a robust scaling method and can be influenced by changes in global distribution.
What do you get with an MA plot. Can you highlight these genes in an MA plot?
Also how about a box plot of log (raw) counts eg
boxplot(log10(counts(dds)+1))
These are the MA plot for design=~sex+genotype and boxplots with
boxplot(log10(counts(dds)+1))