Different FPKM values of cufflinks and cuffdiff in latest version

[mrfox]

Hi all,
This problem has been addressed here http://seqanswers.com/forums/showthread.php?t=4606. And I noticed that in Cufflinks release 0.9.1, the problem of inconsistent FPKM values returned by cufflinks and cuffcompare had been claimed to be solved. Unfortunately, I inspected the FPKM values of many of the reconstructed transcripts and found they have different FPKM values. I am just curious: is it due to slightly different way of estimating FPKM in cufflinks and cuffdiff? Hope this problem will be eventually solved.

Thanks.

[vang42]

I don’t know if this could be the same problem, but I have observed problems with the calculated FPKM values from cufflinks. I don’t know if it is a bug, but the values for a transcript can change, dependent on the size of the GTF file.
I have for example made a small bam file containing only reads from the AVEN gene in the region (chr15:31,945,720-32,118,595 in hg18).
If I run cufflink with a gtf file containing ONLY that gene it will give FPKM of 668896.
If I run cufflink with a gtf file containing AVEN and the overlapping gene CHRM5 it will give FPKM of 655518.
If I run cufflink with a large gtf file containing all or maybe 10.000 genes it will give FPKM of 0.
How can this be?

I attach the relevant files.