FPKM over 10^6

[wjeck]

Hi,

I am using cufflinks for the first time and I noticed an oddity. When I run it on my spliced alignment using a GTF transcriptome file downloaded from ucsc tablebrowser I get a number of genes with an FPKM over 10^6. These genes are longer than 1kb, so this should be mathematically impossible if I understand FPKM correctly.

Has anyone else seen this and know what's going on?

--Will

[endether]

I suspect that you used the -N (upper-quartile normalization) option in cufflinks. Turning on this option will usually result in large FPKM values. Here is the link in the FAQ section of cufflinks explaining why it happens: http://cufflinks.cbcb.umd.edu/faq.html#upquart