Hi,

I performed differential transcript expression analysis using edgeR pipeline. I usually use workflow from the edgeR workflow; 4.6 Differential transcript expression of human lung adenocarcinoma cell lines

It seems it uses the inferential (Bootstrap/Gibbs) replicates to measure the assignment uncertainty of each transcript count. My question is does edgeR also considers transcript length while computing. I see Swish another differential transcript expression tool adjusts transcript replicate counts for sample-specific transcript length and sequencing depth via median-ratio size factors.

Additionally, I would like to know if I can leverage the normalized CPM values for downstream visualization (for instance; to calculate the proportion of transcripts of a given gene)

Here is the related to edgeR approach:

Faster and more accurate assessment of differential transcript expression with Gibbs sampling and edgeR v4. https://doi.org/10.1093/nargab/lqae151

Best Regards, Toufiq

The theory behind the edgeR differential transcript expression (DTE) pipeline is pretty fully described in the paper that you link to. edgeR is also benchmarked against other DTE methods, including Swish, in that paper.

edgeR uses a parametric negative binomial approach, so there is no need to standardize the transcript counts in the way that is necessary for non-parametric method like Swish. edgeR models the variability as a function of read-to-transcript ambiguity and in terms of expected count size, both of which depend on transcript length to some degree, but edgeR has no need to use transcript length directly.

Yes, you can use logCPM for visualization, and edgeR provides functions plotSpliceDGE and plotExonUsage for that purpose.

If you use edgeR for differential transcript usage (DTU), then the plotSplice function can also be used, see

Baldoni PL#, Chen L#, Li M, Chen Y, Smyth GK (2025). Dividing out quantification uncertainty enables assessment of differential transcript usage with diffSplice. bioRxiv. https://doi.org/10.1101/2025.04.07.647659

Note that edgeR does not return TPM. If you want to estimate the proportion TPM expression that each transcript contributes for a given gene, you'd probably need to go right back to the Salmon TPM estimates. All the DTU methods work more in terms of counts than in terms of TPM, because the counts are more statistically tractable. edgeR works with divided counts, which correspond to information content and CV rather than to TPM.