Biological implications of RPKM value

[ThePresident]

Hello everyone,

Currently, I am going deep-through our expression database (bacterial RNA-seq data) in order to get the maximum information possible. I stumbled across very simple but still very important question: From what point can we say that a gene is fairly-well expressed in terms of RPKM value? For example, if I have a RPKM value of 50, with what confidence can I say that that gene is indeed expressed, hence important for the cell in given conditions?

Would it be safe to compare to the mean expression of all genes, or perhaps compare to the expression of some HKG or even tRNA or rRNA?

Thank you in advance for your opinion,

TP

[swbarnes2]

if I have a RPKM value of 50, with what confidence can I say that that gene is indeed expressed, hence important for the cell in given conditions?

Depends on the experiment. My lab throws in ERCC spike-ins into the preps, and if the spike-ins still show appropriately linear behavior at 1 RPKM, then we figure other genes at 1 RPKM are also okay, quantitatively.

In my lab, we usually say that 0.5 RPKM or 1 RPKM are good cutoffs.

An RPKM of 50 means that if you had 1 million reads in your experiment, and you are looking at a 1 kb transcript, you have 50 reads aligning to that transcript. If you have 5 million reads, that's 250 reads on that transcript. That's a pretty solid number, unless your prep has lots of PCR artifacts.

[ThePresident]

Thank you for your input on this.

I plotted a histogram of RPKM value and this is what I got:

https://dl.dropboxusercontent.com/u/7863949/RPKM_histogram.tiff

The median is about 36, but goes from 0 to 140K. So, I guess RPKM of 50 is pretty solid as you suggested, but what I do with low RPKM values? 1 or 2...? Where can I put a cutoff? Is it safe to say that under 10, it is a weak expression, 10-100 is "normal" expression and over 100 is high expression?

Thanks again,
TP