pirl-unc / hitlist / 25069362129
80%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 25
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: 68.718% (+0.06%) from 68.658%
25069362129

push

github

web-flow 
v1.30.0: Load-time allele normalization (#181) + mhc_class_label_suspect flag (#182) (#183)

Two issues confirmed in real data and shipped together because they
share the same code site (load-time post-processing in
``_load_peptide_index``):

1. **#181 — Load-time normalization backstop**. The currently-deployed
   parquet has 19,179 rows from ``source=supplement`` carrying
   unprefixed allele forms (``A*02:01`` instead of ``HLA-A*02:01``)
   because they predate the ``normalize_allele`` call in
   ``supplement.py:118``. ``load_observations`` / ``load_binding`` now
   canonicalize the column at load time using a unique-map over the
   ~hundreds of unique values; the per-call cost is sub-second on the
   full 4.4M-row index. Verified: ``unp.sum()`` goes from 19,179 → 0.

2. **#182 — ``mhc_class_label_suspect`` flag**. IEDB occasionally labels
   short peptides as ``HLA class II`` despite biology (open class II
   groove needs ≥11-13aa peptides). 35K class-II rows in Marcu 2021
   alone are ≤10aa. Add a derived boolean column at load time:
   ``True`` when (length ≤10 AND class II) OR (length ≥18 AND class I).
   Across the full corpus this surfaces 275,995 rows (79K short class II
   + 196K long class I) — surprising tail that's worth manual review by
   downstream consumers.

   Symmetry rationale: canonical class I peptides are ≤14aa with rare
   exceptions; ≥18aa hits are equally biologically suspect.

Both behaviors live behind ``_load_peptide_index`` so every consumer
(``generate_observations_table``, ``generate_binding_table``,
``pmhc_query``, the report) inherits them transparently. No build-time
schema change required — users on stale parquets benefit immediately.

3238 of 4712 relevant lines covered (68.72%)

0.69 hits per line

Coverage Regressions

Lines Coverage ∆ File
1
92.79
 0.64% observations.py
Jobs
ID Job ID Ran Files Coverage
1 25069362129.1 25
68.72
 GitHub Action Run
Source Files on build 25069362129