pirl-unc / hitlist / 25355630698
76%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 28
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 71.434% (-0.04%) from 71.478%
25355630698

push

github

web-flow 
v1.30.36: is_engineered_mhc — narrow is_chimeric to non-native-MHC systems (#226) (#227)

The ``is_chimeric`` flag added in v1.30.34 conflated two biologically
distinct categories:

  - **Engineered MHC**: cells / tissue from one species displaying the
    MHC of another (HLA-Tg rats, NetH2pan training, allogeneic HLA
    transfectants). The presentation machinery itself is non-native.
  - **Heterologous antigen**: host cells presenting a foreign protein
    on their *own* MHC (Lewis-rat EAE with guinea-pig MBP, Equ c 1
    horse allergen on HLA, bovine antigens on murine MHC). The MHC is
    native; only the peptide source is foreign.

Both flag is_chimeric=True because in both ``source_organism !=
mhc_species``. Downstream consumers training "chimeric-aware" models
care about engineered MHC specifically — heterologous-antigen rows are
ordinary host MHC presentation and should bucket with same-species
data, not with HLA-Tg rows.

Implementation:

- ``hitlist.curation.is_engineered_mhc(source_organism, mhc_species,
  host) -> bool``: cached classifier. Discriminator is the IEDB
  ``host`` field — engineered-MHC rows have host genus matching the
  source proteome but differing from the MHC species; heterologous-
  antigen rows have host genus matching the MHC species. Returns False
  conservatively when ``host`` is empty / outside the host-genus
  whitelist.
- ``hitlist.export._compute_is_engineered_mhc``: vectorized helper using
  the same MultiIndex.from_arrays + reindex pattern as is_chimeric, but
  on the (source, mhc, host) triple.
- Wired into ``generate_observations_table`` and
  ``_apply_training_defaults`` next to the is_chimeric block.

Coverage on the v1.30.35 corpus (Class I + II combined):

  - 72,700 rows is_chimeric (unchanged)
  - 69,911 rows is_engineered_mhc (96% of chimeric)
  - 2,789 rows chimeric AND NOT engineered (heterologous-antigen residual)

Top engineered_mhc PMIDs cleanly separate from heterologous-antigen:

  is_engineered... (continued)

3921 of 5489 relevant lines covered (71.43%)

0.71 hits per line

Coverage Regressions

Lines Coverage ∆ File
134
79.98
 -0.94% export.py
34
89.15
 0.34% curation.py
Jobs
ID Job ID Ran Files Coverage
1 25355630698.1 28
71.43
 GitHub Action Run
Source Files on build 25355630698