The tendency for longer genes (and their GO categories etc) to have a greater likelihood to be classified as differentially expressed in RNA-Seq is well known but I'm wondering what people commonly do to account for it? I am aware of a few methods out there (GOSeq, NOISeq etc) but I get the feeling that the most common approach is still to ignore it. Am I wrong in thinking this? One of the most commonly used RNA-Seq toolkits is still the 'Tuxedo Suite' and unless I am mistaken it does not account for length bias in any way.
I should point out that RPKM/FPKM as normalisation methods do not address this problem. Differential expression based on these values is more likely for longer transcripts/genes.
All comments welcome.
FPKM is a normalized unit for describing a rate of observance, but it doesn't address the observation that in evaluating differential expression, longer genes are usually more likely to be classified as differentially expressed. FPKM alone doesn't address this. Bullard et al. (2010) put it this way:
"There is clearly a positive association between gene counts and length, an association that is not entirely removed via scaling by gene length, as in the RPKM" (from "Evaluation of statistical methods for normalization and differential expression in mRNA-Seq experiments", http://www.biomedcentral.com/1471-2105/11/94)
I think they weight their t-statistics by gene length to try to correct for it. In my experience it is largely ignored at this point, and there is no standard approach taken by most people I talk to.
so if it isnt accounted for, shouldnt it be? and what would be the ideally perfect method. Im not clear about why correcting for transcript length wouldnt fix this
I'm no expert... but I think that's the whole point of FPKM?
•
written
by
Brad Langhorst
Log in to answer this question.
