cufflinks-1.0.3 produces very high FPKM values when compared to cufflinks-0.9.3. Why?

[pinki999]

Hi,

The FPKM values produced by the recent cufflinks 1.0.3 are very high when compared to cufflinks-0.9.3. In both the cases I have used -N option. Can anyone tell me whats the reason for this?

Thanks

[chenyao]

the same data and the same reference?

[pinki999]

same samples but previously I had mapped to hg18 build but now to hg19 build. I have two groups with 5 biological replicates. When I used cufflinks-0.9.3 before, the FPKM values between the replicates did not vary much. Now with cufflinks-1.0.3 they are varying a lot.

[Mark.hz]

I encounter the same problem, with -N too. FPKM is pretty high!

[endether]

You may find this link useful: http://cufflinks.cbcb.umd.edu/faq.html#upquart

[drdna]

Cufflinks (v2) makes erroneous FPKM calculations. I first suspected this after looking at the alignment files. Now I have fed the program a control dataset (unique genes, no introns) in which the actual abundances of all "transcript reads" are known. As expected, the FPKM values came out too high. With real data, the most extreme example I've found is where cufflinks produced an FPKM of >400, yet there are ZERO reads from the gene in question. User beware...