rnaseq
rnaseq copied to clipboard
nf-core
Show good / bad examples in documentation
From @Hammarn on March 22, 2017 8:19
It would be useful to supply examples of a good/expected output for all our supported programs. Or we should at least specify wether the supplied results are of a good or bad experiment. FeatureCount is currently shows a library with rather low annotation amounts.
Copied from original issue: SciLifeLab/NGI-RNAseq#93
I don't have example for all the tools on hand, especially those I never used (i.e. previously cited FeatureCount for example) but I'll try to tackle at least past of this. I'd like to at least include a few links to helpful resources like QCfail that can help to sort out the results. If you have examples on hand I'd be happy to add them to the doc as well.
After half an hour of pondering on this problem I'm starting to realise how hard it is to say if examples are good or bad without context. (i.e. are FastQC plots before or after trimming? Is SortMeRNA plot from a rRNA-depleted library? and so on)
At this rate it'll be really hard to provide examples of good/bad output. I think warnings for specific tools would be more appropriate for this pipeline at least. Any thoughts on that?
Adding links where appropriate would definitely be a good start. Yep, it's not trivial to do this sort of thing without maybe having a bad dataset to compare to..
I contemplated working on this issue again yesterday, like maybe launching a few runs of the pipeline on minimal public datasets to generate example plots. However, even if I add good examples only for every step that produces plots, it would amount to a really large number of images on this page, and writing guidelines for all of them seems like a daunting task.
Still, some of the plots currently present in the output docs are showing results that would be worrying in an actual analysis.
What do you think we should do about that @drpatelh @maxulysse ?
I'm thinking let's start just by replacing the bad plots with ones that looks good. Can we point towards the results from the megatests to show what good results should look like?
I'm thinking let's start just by replacing the bad plots with ones that looks good. Can we point towards the results from the megatests to show what good results should look like?
I have no idea when to find those megatests, but if you tell me where to find them I can do that!