To try and catch reads like the R10 10k read S1_46671, where we get two nearly-redundant chains that disagree on ambiguous indel placement, earlier, we could try cutting all the seed anchors down to 1bp, and not excluding chains that would take multiple seeds that claim the same read bases in their minimizers.
This might let the seeds chain up across the disputed indel's location, letting us settle it during alignment generation, and keeping us from needing to consider both possibilities at chaining.
I'm not sure this would produce much better performance than what we do now to drop the nearly-redundant chains, so I'm not sure this is high priority.
adamnovak
commented
2 months ago
To try and catch reads like the R10 10k read
S1_46671, where we get two nearly-redundant chains that disagree on ambiguous indel placement, earlier, we could try cutting all the seed anchors down to 1bp, and not excluding chains that would take multiple seeds that claim the same read bases in their minimizers.This might let the seeds chain up across the disputed indel's location, letting us settle it during alignment generation, and keeping us from needing to consider both possibilities at chaining.
I'm not sure this would produce much better performance than what we do now to drop the nearly-redundant chains, so I'm not sure this is high priority.