Clustering RPKM data

[anjulka]

Hello,

I'm trying to work with RNAseq data that has been pre-processed into RPKM values. I've read a number of threads on here about why that's not the best format, but it's what I have.

I'm trying to cluster the data to identify any underlying groups in the data. A lot of threads suggest using DEseq for all things RNAseq related, but as far as I can tell it requires assigning a sample to a group. Is this understanding correct? I don't want to assign a group; I'd like to identify inherent groups.

Right now, I've taken the data, added 0.00001 to all values, logged (base 2), and standard normalized (subtract mean and divide by standard deviation). But when I try to cluster (using this program - http://code.google.com/p/consensus-cluster/), it's pretty ugly. If I prefilter out all genes that have zero reads in more than 30% of samples, it looks pretty good. But that seems like cheating. Is it?

Is there a better way of doing this?

Thank you.

[Jon_Keats]

though lots of people will have various ideas I'd say you are on the right track, not exactly how I tackle the issue but pretty close and equally valid. I agree DESeq looks very good but not for large cohort as you describe. If you have much microarray experience the filtering strategy you suggest is pretty common and equally appropriate