DEXSeq and DRIMSeq on paired samples?
2
0
Entering edit mode
zoe.ward • 0
@zoeward-16214
Last seen 22 months ago

I have 85 paired samples i.e. before and after ischemia so my question is how would I run the DEXSeq and DRIMSeq on paired samples? I’ve had a look in the relevant vignettes but cannot find any example of paired analysis.

I have a design matrix like so:

 sample_id cond pair 100V_post post 1 100V_pre pre 1 102V_post post 2 102V_pre pre 2 103V_post post 3 103V_pre pre 3 104V_post post 4 104V_pre pre 4 105V_post post 5 105V_pre pre 5

DRIMSeq DEXSeq • 698 views
0
Entering edit mode

Dear Zoe,

One note about the rnaseqDTU workflow is that I initially made an error and included nbinomLRT() in the DEXSeq code. This should be testForDEU(). The workflow code on Bioconductor and on F1000Research has now been fixed to show testForDEU().

2
Entering edit mode
Alejandro Reyes ★ 1.8k
@alejandro-reyes-5124
Last seen 4 weeks ago
Novartis Institutes for BioMedical Rese…

Hi zoe.ward,

For DEXSeq's vignette, this is documented in the section "4 Additional technical or experimental variables". Please let us know if something is not clear. Shortly, you need to pass the following formulae to both estimateDispersions and testForDEU:

formulaFullModel = ~ sample + exon + pair:exon + cond:exon formulaReducedModel = ~ sample + exon + pair:exon

Alejandro

2
Entering edit mode
@gosia-nowicka-9493
Last seen 19 months ago
Basel

In DRIMSeq, based on the design matrix, you would set the full model as

design_full <- model.matrix(~ cond + pair, data = samples(d))

and use it in dmPrecision() and dmFit().

In dmTest(), you can specify which coefficient you would like to test with coef = "condpre" or coef = "condpost", depending on which of the two conditions is as a reference. Or you could define the null model design matrix which would be

design_null <- model.matrix(~ pair, data = samples(d))

This is similar to the case with batches presented in the DRIMSeq vignette section 5.3 Differential transcript usage analysis between two conditions with accounting for the batch effects. In your case, patients are batches sort of.