Deterministic mapping for STAR aligner

[olgabot]

Related from .. a while ago: https://github.com/czbiohub/reflow-workflows/issues/23

Basically, one STAR run to the next is not equal, because multimapping reads are randomly assigned based based on the random seed. If we want the nf-core/rnaseq pipeline to be truly reproducible, should we fix the random seed to e.g. --runRNGseed 0, and maybe expose this as a parameter?

[drpatelh]

Added --runRNGseed 0 6d3e516 to the STAR command here. It makes sense to hard code this for the sake of reproducibility 👍

"will vary from run to run, unless only one thread is used and the seed is kept constant." - although the docs suggest that one thread has to be used too which is very unlikely. Anyway, one aspect is fixed now 😅

[apeltzer]

Maybe my non-native english is in the way here: Doesn't this mean that it will not be deterministic if you set/hardcode it to --runRNGseed 0 and run the process with > 1 thread?

[drpatelh]

Yep. That's my interpretation too. So adding the parameter above won't make a difference unless you use one thread to perform the alignments. But I think having it won't do any harm if our interpretation isn't correct either 😅