Splice variants from a multi factor RNA-Seq

[cfreije]

Hi everyone. We were hoping to find some help using edgeR or DEXSeq to detect/quantify splice variants from a multi factor RNA-Seq experiment.

We have a multifactor RNA-seq experiment that involves two disease states and four time points for each patient (e.g. 4 samples per SAME patient -- basically, a "paired" design with 4 samples instead of 2). We have already successfully completed differential gene expression analysis using the limma/voom pipeline (and edgeR), in which we were able to use the block argument to block the patients and account for this factor in the linear model. We are now interested in looking at differential splicing using either edgeR (spliceVariants function) or DEXSeq. We have summarized reads by exon counts and are ready to proceed with differential analysis, but it's not entirely clear how to pass multiple factors (e.g. disease state, time point, patient) into these workflows.

Can this be done with either tool (edgeR or DEXSeq)? If not, any suggestions for how best to analyze splicing changes in these samples?

Thanks for the help.

[Simon Anders]

Sure, this can be done with DEXSeq.

We are currently working on improving the vignette (manual) to make such things clearer. For now, until we are finished with this, I could try to give you a quick explanation.

However, I am not sure that I fully understand your design: How can one patient have two disease states? If this is not the case, how does blocking help to find the effect of disease?

Maybe, explain what contrasts you used in edgeR, and I can show you how to translate this to DEXSeq.

[cfreije]

Thank you so much for your help.

To clarify:
Our experiment is a time course experiment which is divided into 2 patient groups: disease 1, disease 2. Within each group we follow individual patients for four time points (so each individual has four time points).

Example:
Disease 1:
Patient A Time 1, 2, 3, 4
Patient B Time 1, 2, 3, 4
...
Disease 2:
Patient M Time 1, 2, 3, 4
Patient N Time 1, 2, 3, 4
...

So we are looking for splice differences both within each disease state at different time points (e.g. Disease 1 Time 2 vs Time 1) and across disease states (e.g. Disease 1 vs Disease 2 at Time 2). We would like to take into account the interpatient variability.


Thank you again.

[areyes]

Hi @cfreije,

What you need is to specify the covariates and contrasts in the formula parameters from the function estimateDispersions and testForDEU. Have a look at the part 'Additional experimental or technical variables', from the DEXSeq vignette.

For example, lets say that you want to compare the time 1 vs time 2 of the disease A, you will have to subset your ExonCountSet object and make the design data frame look like this:

patient   disease   time
patientA   diseaseA   time1
patientA   diseaseA   time2
patientB   diseaseA   time1
patientB   diseaseA   time2
patientC   diseaseA   time1
patientC   diseaseA   time2
patientD   diseaseA   time1
patientD   diseaseA   time2

Using the patient factor as a covariate, then you would compare time1 vs time 2 like this:

Your dispersion formula would look then like this:

formuladispersion <- count ~ sample + (patient + time) * exon

and your formulas to test:

formula0 <- count ~ sample + patient * exon + time
formula1 <- count ~ sample + patient * exon + time * I( exon==exonID )


Alejandro