[FEATURE] Call/Confirm SVs with read depth information

[joshuak94]

There are two ways I'm considering for the read depth:

1. We calculate the read depth over the entire BAM file (maybe using a sliding window average), and then use the discrepancies to call breakends. This would occur in the variant detection stage.
2. We feed in a vector of clusters for just insertions and deletions (since those are the only two which can be detected via read depth), and can then either support or not support the call. This would occur after the variant detection and clustering stages.

[Irallia]

Since we assume with 1 that our breakpoints are probably very fuzzy, I would prefer version 2. But we should keep in mind, that there is the possibility of version 1, which we could doublecheck later. Question would be, are there lot of SVs just detected by read depth and how fuzzy are the breakpoints..

In addition to insertions and deletions, there are also duplications (which are basically insertions aswell).

[joshuak94]

I would say using just read depth would be not very accurate. But if you use it in combination with the other methods, then the fuzzy breakends don't matter too much since they'll be supported by things like split reads which have much more precision.

[Irallia]

Okay, then I would say we go with version 1 and then see if our F1 score in the benchmarks suffers or benefits. 👍