RNA-seq: Pathway Analysis - GSVA

[Candace Savonen]

What are the goals of this new example analysis?

We have a GSVA example for microarray, but should create one for RNA-seq.

What kind of dataset will this need?

We need an RNA-seq dataset that we can normalize before running GSVA

What steps should be included in this analysis?

I think most of the GSVA steps from microarray can stay the same with these exceptions I can tell as of now:

• I think we'll need to use DESeq2 and vst() normalized data.
• The Handling duplicate gene identifiers strategy may need to be different (currently its pick the max value for each sample).
• It's unclear to me if/which parameters for the gsva() run in Perform GSVA should be changed -- something to look into.

[Jaclyn Taroni]

Between data types, the main parameter you might want to change is the kcdf argument but if using transformed RNA-seq data it should be the same if I recall correctly.

[Candace Savonen]

When this issue is addressed, note that the intro paragraph from #349 will need to be added here, and the table will need to be made to reflect the RNA-seq versions of the analyses.

[Candace Savonen]

I'm going to try out SRP140558 for this one and see how it goes.

[Candace Savonen]

For handling the duplicate identifiers, for GSVA, I'm not sure mixing values for different Ensembl IDs in RNA-seq makes as much sense as it did for Microarray.

Should I instead switch to something where we pick one Ensembl ID's values over the other(s) -- take them as a set? (Based on bigger average or bigger variance?)

[Jaclyn Taroni]

@cansavvy I would take a look at what was originally in #352 and what is in this comment #352 (comment)

[Candace Savonen]

My guess since you referenced that comment is yes, we still agree think this makes sense since its still per-sample basis?

[Candace Savonen]

A few things we discusssed over video chat that I'm going to change in the draft. These things should help shorten up the notebook (which is currently ~800 lines).

I've taken the general outline from my draft PR and made what are the edits to it:

EDITED:

• Set up data
• Filter out low counts
• DESeq2 normalize and transform (use vst)
• Still Hallmark pathways only
• Do gene ID convert to Entrez IDs
• Resolve multi mapped Entrez IDs same way as before select by a Ensembl ID by max average
• Use normalized data for GSVA
• Differential expression with limma
• Move metadata label cleaning to down here!
• Make sina plot of most DE pathway Heatmap of pathways (without any DE)
• Save plot, session info

[Jaclyn Taroni]

Hallmark pathways only Use all of them!

I disagree with this - there are only 50 hallmark gene sets so you can put them all in a heatmap.

[Jaclyn Taroni]

I'm going to close this issue. Any changes to this example will come about as part of #371.