Soft bed regions v2

[bnoyvert]

The new truvari bench option --extend extends the regions of interest (set in --includebed argument) by the given number of bases on each side, allowing truth set variants to match call set variants that are just outside of the original region. If a call set variant is in the extended regions it can potentially match a truth set variant that is in the original regions turning it to TP, but call set variants in the extended regions that don't have a match are not counted as FP. This strategy is similar to the one implemented in truvari bench for size matching - only the truth set variants longer than sizemin (equal to 50 by default) are considered, but they are allowed to match shorter comparison set variants of size sizefilt (30 by default) or longer.

The default is --extend 0 - no extension, producing the same results as previous versions of truvari bench.

See the following discussion for details.

[ACEnglish]

Looks good. I'm going to change the version back to 3.3.0-dev because we haven't released 3.3, yet. I'll also be shortening the description of the new parameter. I'm probably going to pull most of lines 796:806 to somewhere outside of main as well as probably 814 (that's a long if, might give it a method).

[ACEnglish]

FYI, I made some style changes in 9e4e8a8. Functests all pass.

[bnoyvert]

Thank you, this is all fine.
Just being paranoid - what happens to regions_extended = regions.extend(args.extend) if args.extend else regions if args.extend is a negative number? Would it be safer to put if args.extend > 0?

[ACEnglish]

You already handled that.

truvari/truvari/bench.py

Line 655 in 9e4e8a8

if args.extend > 0 and args.includebed is None:

Edit: nevermind, I stopped reading that line halfway through. I'll think about either fixing there or adding an argparse type requirement of 'positive integer'

Edit2: I'll have to do the argparse type requirement. I realized that extend > 0 isn't the best check because if a user were to specify bench --extend -1 they would silently get results identical to bench [defaults]. Luckily its an easy add-on

[bnoyvert]

Agree, argparse type requirement is probably the best.

[ACEnglish]

See 2f5b001

[bnoyvert]

Just noticed: in bench.py line 647
thresg.add_argument("--extend", type=truvari.restricted_int, default=0,
should be filteg.add_argument. Sorry, my mistake.

[ACEnglish]

It's been changed. I appreciate you double/triple/+ checking. I'm very happy to add your work to the tool. Thanks so much for your contribution!!