Running nextNEOpi on big WGS data

[mantczakaus]

Hi,
I was wondering if you had any recommendations for running nextNEOpi on WGS data where one fq.gz file is over 100GB. nextNEOpi has issues with BWA, MarkDuplicates and ReadBackedPhasing. BWA can take even 18-20h; MarkDuplicates 10-15h and ReadBackedPhasing over 20h. For MarkDuplicates I already modified the code to include sort with sambamba instead of samtools because I read it would be faster. I am attaching my process.config and params.config files.
Best wishes,
Magda
params.config.txt
process.config.txt

[riederd]

In case you are running the pipeline on a HPC cluster or a machine with many cores you could try to increase the scatter count which might help in some places.

e.g. params.config:

scatter_count = "80"

[mantczakaus]

Thanks! Unfortunately, the tasks that take a long time (the three I mentioned) are not parallelised at all so setting the scatter_count to sth higher does not help. Also, because of subdivision-mode of BALANCING_WITHOUT_INTERVAL_SUBDIVISION_WITH_OVERFLOW that is set in gatk SplitIntervals, I end up with 25 intervals only either way.

[riederd]

Hi sorry for the looooong silence, I was very busy last months :-(

You might try to change the options in:

nextNEOpi/nextNEOpi.nf

Lines 741 to 743 in 52d7b92

	--interval-merging-rule ALL \\
	--subdivision-mode BALANCING_WITHOUT_INTERVAL_SUBDIVISION_WITH_OVERFLOW \\
	-L ${IntervalsList} \\

to:

--interval-merging-rule OVERLAPPING_ONLY \
--subdivision-mode INTERVAL_SUBDIVISION \

This would change the size and number (max. scatter_count) of interval lists. However this approach may produce less consistently-sized interval lists so that some scatter jobs will run short and other quite long. But it's worth trying.