pirl-unc / tsarina / 24605667997
74%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 30
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: 33.543% (+2.0%) from 31.522%
24605667997

push

github

web-flow 
v0.6.0: thin iedb.py to a shim, drop broken NetMHCpan subprocess (#7)

* v0.6.0: thin iedb.py, drop broken NetMHCpan shim

tsarina/iedb.py (#5): reduce 481 → 127 lines. scan_public_ms and
profile_dataset become thin wrappers over hitlist.scanner.scan, which
already owns two-header parsing, column resolution, species filtering
via mhcgnomes, source classification, binding-assay detection, allele
resolution filtering, and assay-IRI dedup. peptide_ms_support was dead
and got dropped. _species_kwargs preserves the Homo-sapiens host-field
fallback behavior.

tsarina/scoring.py (#6): delete score_netmhcpan subprocess shim.
Side-by-side validation (HLA-A*02:01, 5 known binders + 3 polyA/K/E
non-binder controls) showed the shim's hand-rolled parser was
silently broken on NetMHCpan 4.2c: affinity values off by ~1000×
and qualitatively inverted (strong binders sub-nanomolar, polyA
scoring like a moderate binder). Topiary/mhctools adapter handles
the 4.2c output format correctly. Net -70 lines, removes a drift
liability the issue flagged up front.

Also includes AGENTS.md Golden Rules section (feature-branch
discipline, version-bump-per-PR, merge-then-deploy, file issues for
blockers, chain foundational work next).

Fixes pirl-unc/tsarina#5, pirl-unc/tsarina#6.

* Review follow-ups: comment _species_kwargs, unit-test it, NetMHCpan breadcrumb

- iedb.py:_species_kwargs grew an inline comment explaining why the two
  branches exist (human_only=True has host/species fallback; mhc_species=
  is strict).  Regression risk: mhcgnomes normalization drift.
- tests/test_iedb.py gets four unit tests pinning the four branches of
  _species_kwargs — catches silent fallback loss if mhcgnomes ever
  changes its canonical species string.
- scoring.py module docstring gets a breadcrumb recording why
  score_netmhcpan was deleted in 0.6.0 — concrete 4.2c parser-drift
  observation so future maintainers don't revive it.

146 tests passing (was 142).

586 of 1747 relevant lines covered (33.54%)

0.34 hits per line

Coverage Regressions

Lines Coverage ∆ File
21
30.23
 9.96% scoring.py
Jobs
ID Job ID Ran Files Coverage
1 24605667997.1 30
33.54
 GitHub Action Run
Source Files on build 24605667997