Feature Request: recursive process

[lindenb]

New feature

as seen on twitter https://twitter.com/yokofakun/status/1248533372630155264 I'd like to have a way to describe a recursive process that would call itself until a condition is done.

I have no idea how it could be described in a DSL:

Usage scenario

• I have a large list of VCF that I want to merge in a given region. The number of VCF is too large
  for bcftools merge or bcftools concat : it would take days and memory to load the indexes and merge the variants. The idea is to merge by divide an conquer the list of VCFs

• same example for BAM

• I want to find the rares variant in my family but absent from 10000 bams:

  • call family-sample1 vs call family-sample2, keep the common variant
  • call family-sample3 only for the variant of previous step, keep the commont variant
  • call family-sample4 only for the variant of previous step, keep the commont variant
  • call family-sample5 only for the variant of previous step, keep the commont variant
  • call control-bams1 remove the variants
  • call control-bams2 remove the variants
  • (...)
  • call control-bamsN remove the variants
    at the end we have the are variants

• GATK CombineGVCF. 10,000 gvcfs to be combined:

  • Combine 100 Gvcf * 100
  • Combine 100 Gvcf
  • Genotype the last GVCFf

Suggest implementation

it's not clear to me how you could implement this idea :-)

recursive process  mergeVcf {
recursionPoolSize 10
stopRecursionWhenPoolSize 1
input:
      val vcfs from vcf_list.andThen(self.merged.collect())
output:
      file("merged${task.recursionLevel}.bcf") into merged      
script:
"""
bcftools merge -O b -o merged${task.recursionLevel}.bcf ${vcfs.join(" ")}
"""
}

[lindenb]

see also http://plindenbaum.blogspot.com/2014/12/divide-and-conquer-in-makefile.html (2014)

Divide-and-conquer in a #Makefile : recursivity and #parallelism.

[pditommaso]

Interesting. I think it could be done using the feedback pattern

[lindenb]

@pditommaso ohh, that's new to me !

[lindenb]

unless I'm wrong your solution would be sequential only ?

In the example you provided, the lines are added one after the other for each process . So, each step is not parallelizeable compared to a divide and conquer strategy (?)

Furthermore, in the scope of merging some BAM files, the main bam would become bigger and bigger and hence, the process slower and slower...

Am I wrong ?

[pditommaso]

Umm, thinking more the collect makes things more complicated and like the loop above does not work because the collect would force to wait for the overall completion.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[pditommaso]

Not sure it's the same feature it's envisioned in this issue, however, the support for recursion has been added recently #2521

[lindenb]

@pditommaso thank you , I'll have a look at this new feature !