Error when validating hgvs_pro deletions: 'AARefAlt' object has no attribute 'ref_n'

[jstone-dev]

When a new score set with hgvs_pro strings includes a deletion, the variants cannot be validated, and the following error is logged:

{
  "type": "AttributeError",
  "exception": "'AARefAlt' object has no attribute 'ref_n'",
  "location": [
    ["/code/src/mavedb/worker/jobs.py", 168, "create_variants_for_score_set"],
    ["/code/src/mavedb/lib/validation/dataframe/dataframe.py", 88, "validate_and_standardize_dataframe_pair"],
    ["/code/src/mavedb/lib/validation/dataframe/dataframe.py", 170, "validate_dataframe"],
    ["/code/src/mavedb/lib/validation/dataframe/variant.py", 169, "validate_hgvs_genomic_column"],
    ["/code/src/mavedb/lib/validation/dataframe/variant.py", 169, "<listcomp>"],
    ["/code/src/mavedb/lib/validation/dataframe/variant.py", 238, "validate_genomic_variant"]
  ]
}

This appears to be due to an unresolved problem in the hgvs package: biocommons/hgvs#727

[jstone-dev]

A bug fix for the hgvs package has been proposed at markgene/hgvs@16682e8.

I tried patching the hgvs library in a similar manner:

lib/patches/hgvs.py

"""
Temporary patches for the hgvs package.

Patches should be removed when the upstream fixes are released.

Patches included:

1. Patch the HGVS package to fix a bug in parsing/validating amino acid deletions.

  The bug is described at https://github.com/biocommons/hgvs/issues/727 and is marked to be addressed in version 2.0 of
  hgvs. This patch is based on https://github.com/markgene/hgvs/commit/16682e80945dba04ce90b5fa814a3ef8527f073f.
"""

from hgvs.edit import AARefAlt

### patches for hgvs issue #727. TODO#672

def get_ref_n(self):
    """
    Return the reference allele length. This property getter is missing from AARefAlt.

    This is a fix for issue https://github.com/biocommons/hgvs/issues/727.
    """
    try:
        return len(self.ref)
    except ValueError:
        return None

# Add a new property to the AARefAlt class. This is equivalent to what the @property decorator does.
ref_n_property = property(get_ref_n)
setattr(AARefAlt, 'ref_n', ref_n_property)

lib/patches/init.py

lib/validation/dataframe/variant.py (inserted line 10)

import mavedb.lib.patches.hgvs  # noqa: F401 TODO#672

However, it seems that AARefAlt is initialized with an empty string as the reference allele, so for single-AA deletions, ref_n is 0 rather than the correct value of 1. This seems to point to a deeper issue in the hgvs library's implementation of HGVS grammar. In hgvs/generated/hgvs_grammar.py, we can see that AARefAlt is initialized with ref="" for deletions. Since the grammar implementation is in generated code, patching it seems risky. If we decide to proceed, the change may be small, but it will require some care.

Otherwise, this issue is tagged for resolution in the future version 2.0 of hgvs, for which there is as yet no target date: https://github.com/biocommons/hgvs/milestone/11.