How to handle limiting call regions

[sorelfitzgibbon]

With the addition of steps to intersect the results of all four of the tools, the question of how to handle different call regions being used by different tools has arisen.

The hg38 reference genome includes the following:

type	labels	number of contigs	total bps
main chromosomes	chr1 - 22, X, Y, M	25	3,088,286,401
alt	_alt	261	109,535,387
random	_random	42	6,978,808
decoy	chrUn_*_decoy	2,385	5,792,522
unplaced	chrUn_	127	4,485,509
HLA	HLA*	525	2,096,467
EBV	chrEBV	1	171,823

Two of the tools, Muse and SomaticSniper, don't have options to limit call regions and so will always call all regions of the reference genome. Mutect2 and Strelka2 both have options to limit the calling to specified regions.

Here are the set of envisioned usage scenarios, and how each would work under the proposed method that does not remove decoy contigs (however see below for a discussion of the much simpler no-decoy proposal).

User runs defaults

• Muse and SomaticSniper will run all regions
• Strelka2 will use main-chr-file to run only those regions
• Mutect2 will use main-chr-file to separate all non-main-chrs to a separate high memory task.
• intersection will use main-chr-file
• general-call-region-parameter = main-chr-file
• no custom regions needed

User provides custom call-regions, e.g. chr1

• Muse and SomaticSniper will run all regions
• Strelka2 and Mutect2 will use the custom regions
• intersection will use custom regions
• general-call-region-parameter = custom regions file

User wants to use defaults except to run Strelka2 on all regions

• Muse and SomaticSniper will run all regions
• `Strelka2 needs just an override to say not to use the main-chr-file
• intersection will use main-chr-file
• general-call-region-parameter = main chrs file
• no custom regions needed

User for some reason wants to run a different custom region on Strelka2 or Mutect2 or for the intersection.

• This seems outside of the scope of the pipeline and would require some additional steps to cater for this at the intersections.
  Instead, they could separately run each set of tools for which they want different regions. If it's two or more, an appropriate intersection would be done.

[sorelfitzgibbon]

Benchmarking of run times on different genome regions - see #201 (comment)

Under the proposal described here, Strelka2 and Mutect2 would by default run on all contigs except decoy, and the intersection would also be done on the same regions. This means the pipeline yields variants for alt and unplaced contigs as well as the main chromosome contigs. There may be a very small increase in average runtime.