openvax / topiary / 24409795081

90%

push

github

Refactor predict_from_variants onto ProteinFragment — v5.4.0 (#121)

PR B of the multi-source antigen unification (#102).

- `predict_from_mutation_effects` builds `ProteinFragment`s from
  varcode effects via `_fragment_from_effect` and delegates to a
  shared `_build_fragment_rows` — one prediction pipeline instead
  of two. Legacy column contract (absolute `peptide_offset`,
  `mutation_start/end_in_peptide`, `contains_mutant_residues`,
  expression-dict plumbing) preserved bit-for-bit.
- Breaking rename (no back-compat alias): `AntigenFragment` →
  `ProteinFragment`, `predict_from_antigens` → `predict_from_fragments`,
  `read/write/iter_antigens` → `read/write/iter_fragments`,
  `docs/antigens.md` → `docs/fragments.md`.
- New regression suites: 75-case frameshift fragment grid pinning
  the varcode `FrameShift.aa_mutation_end_offset ==
  len(mutant_protein_sequence)` invariant; parametrized
  `source_type` vocabulary grid across every
  `_EFFECT_SOURCE_TYPES` entry plus the free-form fallback.
- Underscore-prefixed annotation keys are reserved for internal
  plumbing and never surface as DataFrame columns — documented in
  `docs/fragments.md` and on the `ProteinFragment.annotations`
  field.
- Migration checklist for downstream consumers in CHANGELOG.

96 of 100 new or added lines in 4 files covered. (96.0%)

3006 of 3411 relevant lines covered (88.13%)

2.64 hits per line