Transcript Length Bias in Differential Expression Analysis [DESeq]

[tdyo]

Hi everyone,

Is there a standard approach to addressing transcript length bias in gene expression? I have come across the window approach provided by Oshlack & Wakefield (2012), but not much else.

I have run Bowtie2 with insert restrictions, ignoring discordant alignments, and removed reads that are not uniquely aligned afterwards (resulting in a uniquely aligned, concordant SAM file). Following this, I ran the SAM file through htseq-count and DESeq. The following graph shows the significant relationship between scaffold length and the BaseMean Expression provided by DESeq (the p-value is very low, but the R2 is also very low). Any thoughts? Thanks for any input!

[chadn737]

What exactly are you trying to do? You are asking a very general broad question with no clear indication as to why transcript length is a factor in your analysis. For instance if you are simply comparing the expression of the SAME gene under two different conditions (the standard usage of DESeq) than transcript length is not a factor since you are comparing transcripts of the same length.

[tdyo]

Yes, I am comparing the expression of the same transcript to itself between two groups of individuals under different treatments (looking for differentially expressed genes)... well that makes sense. Now I feel silly - thanks for the insight.