Awesome new package! Feature question: are you going to add result convertors?

[lcolladotor]

Hi,

I love this new package! It should make it very easy to then run both DESeq2 and edgeR.

As you can see at:

Awesome new @Bioconductor package by Andrzej! Any plans on adding convertors for DESeq2 and edgeR results? https://t.co/mFr5R0nsFz

— L. Collado Torres (@fellgernon) March 25, 2016

I'm curious if you plan on adding convertors for switching the results from one package to the other one. This is certainly much harder. I've tried a bit at https://github.com/leekgroup/derSupplement/blob/gh-pages/simulation/deseq2-edger/calc_stats.R to make the results look kind of similar, but you don't get the same metadata columns. You can find some of the Rdata files at https://github.com/leekgroup/derSupplement/tree/gh-pages/simulation/deseq2-edger

Anyhow, this is not a feature request, just me being curious. I don't know how far you plan to go with DEFormats.

Best, Leo

PS Couldn't find you on Twitter.

[aoles]

Hi Leo, many thanks for your feedback! It's great to hear that you find my package useful.

My primary focus was on providing common entry points for various DE calling pipelines. In my initial implementation I didn't plan to solve the problem of converting between the results. The main reason is, as you already noticed, the differences in the output, and the resulting lack of 1:1 correspondence between the objects.

Thank you for sharing your attempts to address this issue. I might consider extending the converters to post-analysis objects in some future updates, so stay tuned.

Best, Andrzej

PS: I'm twitting under @andrzejkoles

[lcolladotor]

Sounds good Andrzej!

Also, this paper and the package behind it might be relevant to DEformats http://biorxiv.org/content/early/2015/04/17/018200 I would also check http://nar.oxfordjournals.org/content/42/11/e91 and in particular http://imlspenticton.uzh.ch/robinson_lab/edgeR_robust/ which includes files such as http://130.60.190.4/robinson_lab/edgeR_robust/robust_simulation.R

One idea is to have DEformats have p-value extractors from DESeq2 and edgeR results (among others), and then use benchmarkR to make some plots if you have a truth set.

[lcolladotor]

Hi @aoles,

You'll notice that regionReport now imports DEFormats. That's mainly due to the new function edgeReport() where I have a manual result convertor from edgeR to DESeq2 results. See https://github.com/leekgroup/regionReport/blob/72a05851793a7b40fbec6cb449a2948c3d784d83/R/edgeReport.R#L113-L147 for details. The conversion is not complete, but it is good enough for the purposes of what I need in regionReport.

If you end up creating more extensive result convertors, I'll use them. It's code that should most likely live in DEFormats instead of regionReport. Meanwhile, this is a solution that works for my use case.

Best, Leo

[lcolladotor]

I forgot to tag @mikelove with whom I've talked a bit about this before.