Significance of an FPKM value

[JueFish]

Curious what some other people might be thinking about what I am seeing in the literature these days. Am I wrong that the general contention that people make that an FPKM value = 1 (estimated in Cufflinks) is effective equivalent to 1 transcript per cell erroneous as that value of FPKM is determined by an variety of factors, not the least of which is the size of the library used in the analysis and/or how one choose to weight their FPKM values? I see FPKM as a strong relative measure for comparisons (due to all the business going on with cuffdiff and cuffcompare), but an unreliable absolute measure without experiment specific validation. Is that a fair statement? I thought I'd ask around as I often see statements along this line of thinking for validating a threshold cut-off for data analysis but as far as I can its related to one study, and should hardly be the broadly applied "standard" it seems to be.

[turnersd]

If I understand this correctly, the absolute levels of a transcript can only be calculated when you have experimental mRNA standards that you sequence, or if you have a robust estimate of the mRNA content of the cell type you're looking at. From the 2008 Mortazavi paper:

To assess the dynamic range of RNA-Seq and to test for possible effects of starting transcript length on the observed transcript abundance, we introduced into each experimental sample a set of known RNA standards transcribed in vitro from Arabidopsis and phage lambda templates... When these RNA standards are used in conjunction with information on cellular RNA content, absolute transcript levels per cell can also be calculated. For example, on the basis of literature values for the mRNA content of a liver cell19 and the RNA standards, we estimated that 3 RPKM corresponds to about one transcript per liver cell. For C2C12 tissue culture cells, for which we know the starting cell number and RNA preparation yields needed to make the calculation, a transcript of 1 RPKM corresponds to approximately one transcript per cell.

[JueFish]

Thanks for your thoughts, turnersd. I got an email back from Cole Trapnell and Lior Pachter about this issue and it was pretty much as I thought. They seem to be trying to work on a bioinformatic solution to this, which would be interesting. In my experience, those FPKM values can fluctuate from library to library and it's driven by a variety of factors (Iibrary size, patterns in transcript abundance, etc.). I actually was referring specifically to papers that cite the Mortazavi et al 2008 paper and use it as a means of filtering data for their own analysis. This seems to be a questionable approach to me, unless you were dealing with a similar library type (25bp, Illumina, 42-51 milllions reads, etc) and tissue type. It underscores two points for me: 1) you need to have some means of validating your FPKM values (qPCR, spike-ins, etc.) to say anything about absolute abundances; 2) reviewers need to be more vigilant about how data filters are applied in manuscripts as the application of one cut-off threshold from one unrelated experiment may be largely meaningless to the context of another.

[turnersd]

I'm trying to rationalize what a bioinformatics/analytical solution to the X FPKM=Y #transcripts problem. Did Cole/Lior share any interesting thoughts?

[JueFish]

Lior only said they were trying to think of a solution to this issue, but haven't come up with something satisfactory yet while trying to explore some possibilities using transcript coverage estimates. It would be interesting to see if there is a computational solution, but for now I'm much more comfortable with my spike-ins as they can help account for all type of technical errors during library construction as well. Any informatic solution is still likely going to be insensitive to the technique of the individual constructing the library and doing the sequencing.

[wupengpro]

The same problem also bothers me. Who can provide some detailed suggestion about how to choose appropriate FPKM cut-off to judge whether the corresponding gene is expressed?