DiffBind - suggestions for modifications

[mblue9]

I was thinking of making more changes (below) to the DiffBind wrapper (as I can get the time). @zfrenchee as your group has made requests previously for modifications to this tool are you (or anyone else) interested in any of these changes? As I could prioritize changes higher that other people would like too.

• [ ] 1. add a test using control BAMs (need to create downsized control BAMs to match the test-data data, but note that control BAMs are not really recommended for the diffbind step)
• [x] 2. output Rscript like EGSEA
• [x] 3. output the dba.analyze "log" to show what was used (e.g. DESEQ2 method) and maybe output the FrIP scores as a separate table (left output as just a text file, not table)
• [x] 4. output more of the diffbind plots e.g. PCA, MA, Volcano, Boxplots (only correlation heatmp currently produced)
• [ ] 5. allow multiple contrasts to be specified like limma-voom/edgeR (at the moment only 1 is enabled)
• [ ] 6. maybe allow contrasts (and their order e.g. A-B or B-A) to be specified in the tool form like limma-voom/edgeR
• [ ] 7. allow less samples to be used, currently requires 4, could enable 3 or maybe 2
• [ ] 8. add parameters e.g. for
• [x] - scorecol
• [x] - summits
• [ ] - method deseq2/edgeR
• [ ] - blocking

[bwassie]

@mblue9 These modifications would be great! Just so I understand, item 2 will create an Rscript we can use with edgeR/DESeq2? If so, then I would be very interested in items 2, 4, 3, 7, 8 in that order. More plots would be nice; we usually look at the PCA plot to see if our samples cluster and the MA plot would be nice to visualize the normalization. The other items I mentioned would also be helpful for us! Also, the DiffBind tool can take collections now if I'm reading the linked thread correctly? If so this would be of great use to us!

[mblue9]

Thanks @bwassie! I'll see what I can do. For 2 I meant output the Rscript used by the diffbind wrapper so people could see what commands are being run like here: https://github.com/galaxyproject/tools-iuc/blob/master/tools/egsea/test-data/out_rscript.txt (I now see I never added an rscript output to limma-voom/edgeR but I've done it for other tools like egsea). Although, as diffbind is using deseq2/edgeR for it's stats, maybe it could integrate a bit with the Galaxy versions of those tools, maybe you could feed the diffbind normalised counts/binding matrix in...hmm need to think about that.

[mblue9]

I discovered with diffbind that you can make a funky 3D PCA plot that you can twirl around (fun :dizzy: ) and it also looks it could be useful for explaining PCA dimensions (something I find a lot of people struggle with). No idea if it's possible to implement this in diffbind Galaxy but if anyone's interested to have a look you can generate the 3D plot with:

library(DiffBind)
library(rgl)
load("DiffBind_analysis.RData") #this is in the diffbind test data here: https://github.com/galaxyproject/tools-iuc/blob/master/tools/diffbind/test-data/DiffBind_analysis.RData
dba.plotPCA(sample_analyze, b3D=T)

[pavanvidem]

One more request from a user is to add minOverlap argument as a parameter in the wrapper. This is used in different functions of DiffBind. In addition to the vignette and reference manuals, there are some hints from the tool author here and here.