Running multiple instances of STAR in one directory

Question

[Michael]

We are trying to integrate STAR into a galaxy pipeline. As you might know, STAR can only run as a single process per directory, which is currently the main obstacle. The main reason for star not being able to run in parallel is that names of the output files are not parameters but fixed. The program will refuse to run even when detecting an already running star in the directory.

Is there a plan or a way to make the output file names configurable using a prefix so it will not fail? Maybe a source modified version with this feature?

Second best option, to give each star its own chrooted tmp environment, but I would like to go for the cleaner option first.

Answer 1

[Devon Ryan]

I've gotten around this on our my old cluster by giving each instance a different --outFileNamePrefix, does that not suffice for you? This does still create some annoying empty temp files in whatever directory you execute STAR in that need to then be removed (they don't block multiple instances though).

[Michael]

My bad, I must have totally overlooked this option. Thanks a lot! Does it also work with genomeGenerate?

Typical RTFM case....

[Devon Ryan]

I've never tried this with genomeGenerate, but probably not. I think you need to run things in different directories there.

[Devon Ryan]

BTW, Alex Dobin has always seemed really amenable to modifying STAR, so he'll probably add an option for this if there isn't already one.

Also, the iuc already has a STAR package in Galaxy. Does that not suffice for making a workflow out of things (truth be told, I only admin our Galaxy server, I don't personally use it)?