DESEQ2 versus Cuffdiff

[David Lyon]

I have 2 RNAseq samples before and after treatment to give a total of 4 files of FastQ
When I look for DEG using Cuffdiff I get different answers compared to using DESEQ2.

The DEG set are mostly different with min overlap.
Raw counts from HTSeq are different from FPKM.

So which set of results do you use, the cuffdiff or the DESEQ2 results ???

[student-t]

The answer is it depends. Note that they use a very different statistical model, so it's not fair to compare the p-values. You'll have to understand the statistical model before you can conclude anything.

No package can give you a definite answer because it only works if you believe in the model. If I started a new package and used a simple independent t-test, would you believe me just because that the p-value is significant? No, you'd conclude my modelling assumptions were wrong.