On Wed, Oct 3, 2012 at 9:11 PM, i b ibseq12@gmail.com wrote:
Dear all, how reliable is running Cuffdiff without replicates? e.g.one samples agains another one?
Is it statistically makign any difference when using replicates?
Seqanswers might be a better place to ask this very interesting technical question that goes way beyond Galaxy...
My 2c: Statistically speaking, sequencing and biology are both noisy. Replicates provide information about non-experimental (technical and biological) variation. That variation is usually not the variation you are looking for, but if you want to remove it, you have to model it and that requires information from replicates (or really good guesswork). In some situations (eg extreme experimental conditions), I'm sure you'll find biologically meaningful signal without them but in my experience, they can really help to decrease non-experimental noise, particularly where the experimental condition induces only subtle changes in transcript abundance.
You could always analyse a data set with replicates and compare the results with and without those replicates yourself to see what happens - it would be a nice paper I'm sure.
Thanks, ib