[miso-users] cluster job memory question

[Yarden Katz]

Hi John,

I believe, based on a previous user's report, that this might happen when the AFE annotation for an event marks an extremely large region.  In that case, the annotation is probably wrong, but the consequence is that potentially hundreds of thousands of reads are loaded into memory by MISO.  Let me know if this explanation makes sense with what you've seen.  Version 1 annotations for hg19 can have these misannotation errors, in part because they are made by liftOver (from hg19 -> hg19) which is imperfect.  

In any case, I will add a fix for this in an upcoming release.  If you do have a particular set of events that you know for sure caused this on your system, it'd be helpful if you can send those along.  A quick temporary fix is to filter out extremely long events (based on their GFF coordinates) from AFE 1.0 hg19, reindex, and re-run MISO.

Best, --Yarden


On Feb 23, 2014, at 12:00 AM, John Tobias wrote:

> Hello,
> 
> I’m using miso (0.5.0) on an LSF cluster to compute alternative events in an exon-centric analysis. I find that some jobs are taking more than 64GB of memory and exiting (that’s the limit on my cluster). I see this using the AFE 1.0 hg19 gff3 annotations. Is there a way to adjust miso settings such that I can limit the amount of memory used?
> 
> Thanks,
> 
> John
> _______________________________________________
> miso-users mailing list
> miso-users at mit.edu
> http://mailman.mit.edu/mailman/listinfo/miso-users

--
Yarden Katz <yarden at mit.edu>