Differential Transcript Usage: complex factorial design
2
0
Entering edit mode
jbono ▴ 10
@jbono-7682
Last seen 23 days ago
United States

Hi all,

I would like to analyze differential transcript usage (DTU)/Isoform switching following Salmon quantification, but have a fairly complex three-way factorial design. I am interested in three factors: Genotype (mutant, WT), Sex (M, F) and Age (1D, 28D). I am not only interested in some simple contrasts, but also in some interaction contrasts (e.g. Genotype by Sex and Genotype by Age). I used edgeR for DE analysis and was able to pretty easily set up the appropriate contrasts, but, at least from my understanding, edgeR is not really set up to perform DTU analysis. I would ideally like to do something similar for DTU rather than performing all pairwise comparisons separately and not really being able to look at interactions. However, I am getting stuck finding an analysis pipeline that would work with this complex design. I've considered DEXSeq and IsoformSwitchAnalyzeR but it is not clear to me from the documentation how to set up the contrasts I would be interested in (it seems like maybe it's possible but I'm not sure). I would greatly appreciate any advice on whether these or other packages would work for this design, or whether pairwise comparisons would be the best way to go. Thank you in advance for any advice!

edgeR DEXSeq IsoformSwitchAnalyzeR DTU • 176 views
1
Entering edit mode
@gordon-smyth
Last seen 42 minutes ago
WEHI, Melbourne, Australia

edgeR does differential expression for transcripts via the catchSalmon function followed the usual edgeR pipelines as you would do for gene-level DE. Inputing the same counts to diffSpliceDGE will then analyse DTU.

edgeR detects differential exon usage and differential exon-junction usage via featureCounts and the diffSpliceDGE function, which is a statistically more powerful approach than DTU.

0
Entering edit mode
Alejandro Reyes ★ 1.8k
@alejandro-reyes-5124
Last seen 2 days ago
Novartis Institutes for BioMedical Rese…

Hi jbono. It is possible to set up these complex interactions in DEXSeq. As a quick example, if you are interested in genotype:sex contrast, the contrasts would be:

formulaFullModel    =  ~ sample + exon + sex:exon + genotype:exon + age:exon + sex:genotype:exon
formulaReducedModel =  ~ sample + exon + sex:exon + genotype:exon + age:exon


Alejandro

0
Entering edit mode

Thank you both for this useful information, I really appreciate it!

0
Entering edit mode

Hi Alejandro,

I noticed your suggested full model for testing the genotype:sex interaction does not include an age:sex interaction. Is there anything wrong with leaving the age:sex interaction in the full model statement? Also, to test main effects, can I just use the full model with all the terms and their interactions and then drop each main effect (e.g. genotype:exon to test for genotype) in the reduced model, or is it more complicated than that due to the presence of interactions? If so, would I instead just do: ~sample+exon + sex:exon + genotype:exon + age:exon for the full model and drop each factor one at a time in the reduced model? Finally, if I am interested in the simple contrasts (e.g. control vs. mutant females at age 1) would I need to subset my data to include only those samples, doing this separately for each contrast, or is there some way to fit the full model and test individual contrasts? From the manual it doesn't seem like that is possible, but I just wanted to check. I appreciate your time and insight, thanks!

0
Entering edit mode

I don't see anything wrong leaving age:sex if you are interested in that interaction. And I think you are right in leaving all the terms in the formulate and drop the interaction term that you are interested in the reduced model. For more specific contrasts, like control vs mutant only in females, the easiest way is to subset the data and rerun both estimateDispersions() and testForDEU().