analysis: re-run civic/moa variant analysis + add zip for moa features to reuse

[korikuzma]

There are some VRS ID changes. Here's an issue I made for gnomad vcf queries in the variation normalizer. I'm going to keep this as a draft PR for now to investigate why the normalization rate went down

[korikuzma]

@wesleygoar Some reasons why the normalization rate dropped is because our regex is case sensitive in the refactor. Here are some examples from CIViC that fail due to case sensitivity:

E588_Y589INSKYFYVDFRE
M552_W557DEL

We added case sensitivity to make it clear what the variant passed. For example, if E588_Y589DELINSKYFYVDFRE was passed would this be a deletion (with invalid deleted sequence at the end) or a delins?

[korikuzma]

@wesleygoar we dropped in both civic and moa but not by much (see analysis files for percent changes), so let me know if you want me to merge these changes or if you'd like us to come up with solutions in the normalizer refactor.

[wesleygoar]

@korikuzma The VRS ID changes were due to the variant being valid in both builds?

[korikuzma]

@wesleygoar Yup

[wesleygoar]

Yeah we should fix that.

[korikuzma]

I know, we discussed me solving the linked issue before via slack

[wesleygoar]

I know. ; ) Just doing my part to keep provenance via github like you wanted.

[ahwagner]

I think we can resolve this without case delineation. For any given expression of this form, there can only be:
1 valid interpretation
0 valid interpretation

For example, E588_Y589DELINSKYFYVDFRE is not a valid deletion (as Kori pointed out) but could be a valid delins, and therefore we would resolve as DELINS.

If a string has no valid interpretation, then it is simply invalid and we move on.

[korikuzma]

@ahwagner What if we had a case like I588_I591DELINSI. Where I does exist at pos 588 and I does exist at pos 591. This could be a deletion of 588-591 where the sequence is INSI. Or it could be a delins where 588-591 is deleted and I is inserted. Would we want to use delins or deletion?

[ahwagner]

Yeah. Wes and I were discussing this down here while you typed up this example. This is part of the narrow class of n-3 length expressions where we have this challenge. I would say we just attempt to interpret both ways, and if it is ambiguous we say we can't normalize due to ambiguity.

[korikuzma]

Yeah we should fix that.

@wesleygoar PR is made here.

[korikuzma]

@wesleygoar re-ran with input_assembly parameter for civic.

[wesleygoar]

@korikuzma thanks. Ill take a gander at the data.