Fix tx_filter_variants_by_csqs to correctly handle the ignore_splicing parameter

[jkgoodrich]

As is, the ignore_splicing parameter will remove transcripts from transcript_consequences where the worst consequence in "consequence_terms" is a splicing consequence. However, we want to keep the transcript if it has another consequence in "consequence_terms" that is not a splicing consequence. This PR updates the function so that splicing consequences are removed from the "consequence_terms" before the most severe consequence is added.

Example:

 "transcript_consequences": [
    {
        "consequence_terms": [
            "splice_region_variant",
            "synonymous_variant"
        ]
    }
]

Currently this transcript would get removed, but we want it to be kept as:

 "transcript_consequences": [
    {
        "consequence_terms": ["synonymous_variant"],
        "most_severe_consequence": "synonymous_variant",
    }
]

[ch-kr]

a couple questions

[ch-kr]

I wonder if this code should actually be added to the VEP utils? my initial thought was to put it here, but it doesn't quite fit with that function

[ch-kr]

also, does this mean variants that had a splicing consequence as their worst consequence were also removed from the previous pext results? I found this line in the original repo https://github.com/macarthur-lab/tx_annotation/blob/master/tx_annotation.py#L305 (but didn't dig thoroughly)

[jkgoodrich]

I actually started to add it to the vep utils, but then decided against it because that was clearly filtering based on most severe consequence, and I thought it might be confusing to have both.

[jkgoodrich]

This fix is to make it do what was done in the original. The line of code you reference picks the most severe consequence after removing the splice consequences from the ordering

[ch-kr]

yeah, I keep going back and forth on whether this (filtering on any csq) is helpful to have, but maybe you're right that it might be too confusing

[ch-kr]

LGTM!