DTU using DEXSeq (time series data with condition factor)
0
0
Entering edit mode
@261fa811
Last seen 7 months ago
United States

Hello, My setup looks like the following:

sample   timepoint        condition

s1          0               plus
s2          0               plus
s3          0               minus
s4          0               minus
s5          1               plus
s6          1               plus
s7          1               minus
s8          1               minus
s9          2               plus
s10        2               plus
s11        2               minus
s12        2               minus


I am trying to set up a DEXSeq DTU analysis. I am following this protocol: https://f1000research.com/articles/7-952/v3

So the model is going to be interaction between time and condition. How do I set this up using Dexseq, but at the DTU level? I am not sure what the full or reduced model will be.

Dexseq already requires an interaction term, which is condition:exon. I guess the interaction that I am interested in is condition and timepoint.

This is easier to do using DRIMSeq, as the full model will just have condition:timepoint, with the reduced model not having this interaction.

After thinking this through, I think the models will be:

formulaFullModel    =  ~ sample + exon + condition:exon + timepoint:exon+ condition:timepoint:exon
formulaReducedModel =  ~ sample + exon + condition:exon  +timepoint:exon


However, I am wondering though, since we are comparing different isoforms, that the following reduced model makes more sense. Any thoughts?

formulaReducedModel =  ~ sample + exon  +timepoint:exon


For example, if the usage of an isoform goes from 30% to 40% to 50%, I think that baseline is also important.

Would appreciate if somebody could verify. Thanks in advance.

DTU DEXSeq RNADTU • 481 views
0
Entering edit mode

Apologies for late reply, Nirad. You probably figured it out by now. In the formulae, the 'exon' term would take the baseline transcript usage. The 'condition:exon' and 'timepoint:exon' would capture the changes of condition and timpoint vs the baseline, and 'condition:timepoint:exon' the interactions of condition and timpoint in exon usage. I think you could vary the reduced model depending on the effects that you are interested in capturing in the test.