pirl-unc / hitlist / 25268809988
80%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 27
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: 71.36% (+0.4%) from 70.97%
25268809988

push

github

web-flow 
v1.30.30: Sample-resolver tie-break by cell_name + obs metadata (#219)

Two fixes that together unlock ~165k newly resolved sample-level
peptides across 5+ APM perturbation studies (#207, #209-style ask
from APM aggregation).

1. Normalize YAML allele tokens via mhcgnomes before joining.
   Previously curators wrote `H-2Kb` while IEDB stores `H2-K*b`,
   so the resolver's exact-string match silently missed every mouse
   allele. Same fix covers any non-canonical syntax that mhcgnomes
   knows. PMID 34428180 (Son 2021): all 70K rows used to fall to
   pmid_class_pool; now 58K allele-match.

2. Per-obs-row tie-break when multiple ms_samples share an
   (pmid, allele) or (pmid, mhc_class) key. Previously the resolver
   collapsed them onto an arbitrary first-pick, so e.g. all mouse
   hepatocyte / spleen / skin rows landed on whichever sample
   appeared first in YAML. Now each obs row scores candidates by
   token-prefix overlap between sample_label/perturbation and the
   IEDB row's cell_name / source_tissue / antigen_processing_comments
   / assay_comments, with rarity weighting (tokens unique to one
   candidate count more than tokens shared by all) and word-boundary
   substring containment. A per-(pmid, class) variance check drops
   constant study-level fields from the tie-break input so they
   don't dominate the rarity score.

Concrete impact (sample-resolved peptide counts, MHC-I-relevant
APM perturbations only):

  intervention      before        after
  IFN-gamma             44       81504
  ERAP1              15087       61170
  ERAP2              16127       32137
  Mouse TAP1             0       16665
  Tasmanian devil
    DFT1+IFN-γ           0       13731
  TNF-alpha              0        4676

Tests: 7 new unit tests in test_export.py covering rarity weighting,
word-boundary substring (the "treated"-in-"untreated" trap),
empty-obs fallback, single-candidate fallback, mouse normalization,
and the hepatocyte/spleen split. Full suite: 610 ... (continued)

3847 of 5391 relevant lines covered (71.36%)

0.71 hits per line

Coverage Regressions

Lines Coverage ∆ File
120
81.95
 0.52% export.py
Jobs
ID Job ID Ran Files Coverage
1 25268809988.1 27
71.36
 GitHub Action Run
Source Files on build 25268809988