identify high and low express genes from cufflinks output

Question

[hana]

Hi

I want to find high and low express genes from cufflinks output (genes.fpkm_tracking file) in one sample. How can I choose an FPKM cut-off to judge whether the corresponding genes are High expressed or not?

There are some papers that use 1<FPKM . By using this threshold I found ~ 11,000 genes, but I would like to find most high and low express genes.

I would be grateful if you could give me suggestion.

Thank you

Answer 1

[Devon Ryan]

Make a histogram of the FPKMs and see if there are multiple humps. If so, you have your threshold. If not, you get to arbitrarily pick a value that looks nice (I would personally just take the top x% as highly expressed). The FPKM<1 that you see in some papers is just a nice round number with no actual legitimacy (regardless of what anyone might say to the contrary).

[hana]

thank you so much

Answer 2

[Charles Warden]

There are some statistics for a few cutoffs in this paper.

That said, the goal is a little different: I think 0.1 is generally a good threshold to exclude low coverage genes that will tend to have high fold-change values, but I wouldn't necessarily describe genes with > 0.1 RPKM as having "high" expression (rather I would probably just say they are "expressed")

[hana]

Thank you for your reply