openvax / topiary / 33402884781

93%

Build Type

push

github

Committed by web-flow

Commit Message

Classify alleles with mhcgnomes instead of a name prefix (#209)

derive_mhc_class read HLA-A/B/C as class I and HLA-D* as class II, so every
other real allele came back pd.NA: the non-classical human class I genes
(HLA-E, -F, -G) and all non-human MHC (mouse H2-Kb and H2-IAb, BoLA, Mamu,
SLA).

pd.NA is not a harmless answer. It drops a row from the class_i and class_ii
filters alike, so those peptides were in neither view rather than in the wrong
one - the same silent-absence shape as the other defects fixed this cycle.

AGENTS.md has said this all along: "Use mhcgnomes for MHC allele parsing. Never
startswith("HLA-") or other string hacks - alleles aren't always human." This
was the one place in the codebase not following it.

The class comes from mhcgnomes' own is_class1 / is_class2 / has_mhc_class
predicates rather than from string-matching its Ia / Ib / IIa / IIb labels,
which would be the same guess one layer up. Distinct alleles are parsed once
per call.

Claude-Session: https://claude.ai/code/session_01BAUqjqDDPXaTdtUDJqapJG

Coverage Stats

16 of 16 new or added lines in 2 files covered. (100.0%)

1 existing line in 1 file now uncovered.

5518 of 6073 relevant lines covered (90.86%)

2.73 hits per line