Cuffdiff stranded and unstranded data

[jake13]

I have generated stranded RNA-seq data and have been comparing it using cuffdiff. However, I now want to compare some of my samples to a previously published data set that does not have stranded data. Can I compare these samples in cuffdiff? If so, what do I specify for the library type?
Thanks

[Brian Bushnell]

No, you can't really do a valid comparison unless the organism does not have overlapping opposite-strand genes. Fortunately... outside of prokaryotes, organisms generally don't have overlapping opposite-strand genes. So if you're working with multicellular eukaryotes, just declare both libraries as unstranded and compare the results directly. Try evaluating the same library against itself, processing it stranded versus unstranded. If the results indicate no difference, you might be OK. The different protocols may have different biases, though. I would consider the data as informative, but I would certainly not publish anything based on comparing datasets with different methodologies.

[wenhua]

cuffdiff for DEGs

I use tophat ,cufflinks and cuffdiff analysis of RNA-seq 2 samples of three biological replicates .
I want to know why the value of log2(fold_change) between two samples is relatively large(such as 6) and q>0.05 resulting in significant value is "no".
How to use "log2(fold_change)"and "q value" in screening differentially expressed genes between the two samples.

Another case is that one gene FPKM value in sample1 is 0 and,in sample2 is greater than 0 (such as 14), the result from cuffdiff also no difference between two samples.Why?
Anyone else notice this?

Thanks

[jwfoley]

You could just accumulate per-transcript read counts and use a more flexible program like DESeq2.

Of course there is the risk of ambiguity for overlapping genes, as Brian Bushnell points out (and partial overlap does happen frequently in multicellular organisms), but if you're worried about that then you have to discard your unstranded data altogether.