pirl-unc / tcrsift / 26638989809
84%

DEFAULT BRANCH: main
Ran
Jobs 4
Files 31
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: 77.296% (+0.005%) from 77.291%
26638989809

push

github

web-flow 
2.0.1: fix #107 — validator was rejecting junction-prepended constants (#110)

Regression in 2.0.0. ``validate_sequences``'s canonical-start check
sourced the expected canonical from ``HUMAN_CONSTANT_REGIONS_AA[base]``
— the bare-mature canonical without the per-clone junction residue.
But after #105, the row's ``{chain}_constant_aa`` carries the
junction-prepended layout (e.g. ``NIQNPDPAVY…`` for α with N junction,
``EDLNKVFP…`` for β with universal-E junction). The observed start
extracted from ``full_{chain}_aa`` past VDJ included the junction;
the expected didn't; every clone failed.

Effect on users: ``tcrsift run --contigs-dir`` exited non-zero at
the PDF gate (~4500 false-positive load-bearing failures across two
audited donors) even though all data outputs were correct.
``strict=False`` / ``strict='skip'`` worked around it.

Fix: validate_sequences now prefers the row's actual
``{chain}_constant_aa`` over the bare-mature canonical lookup. The
row's per-clone value already reflects whatever junction logic ran
in ``_add_constant_regions``, so observed and expected are
byte-for-byte comparable. Falls back to
``HUMAN_CONSTANT_REGIONS_AA[base]`` when the row doesn't carry the
per-clone constant (rare; preserves pre-2.0 behaviour on hand-built
frames).

Tests:
  - ``test_validate_sequences_accepts_junction_prepended_layout``:
    E2E fixture with contigs loaded; assert
    ``create_tcr_sequence_pdf`` / ``validate_sequences`` emit zero
    "constant start doesn't match canonical" messages for either
    chain. Would have caught #107 on day one — the gap was that 2.0
    only exercised assemble.py unit tests, not full-pipeline E2E
    with contigs through validation.
  - ``test_validate_sequences_still_catches_real_start_mismatch``:
    defensive — when ``full_alpha_aa`` doesn't match its own
    ``alpha_constant_aa`` after VDJ, the validator still flags it.
    Locks the fix from softening too far.

Suite: 1218 passed (+2); ruff clean.

6564 of 8492 relevant lines covered (77.3%)

3.09 hits per line

Coverage Regressions

Lines Coverage ∆ File
17
90.65
 0.02% assemble.py
Jobs
ID Job ID Ran Files Coverage
1 python-3.11 - 26638989809.1 31
77.29
 GitHub Action Run
2 python-3.10 - 26638989809.2 31
77.29
 GitHub Action Run
3 python-3.12 - 26638989809.3 31
77.29
 GitHub Action Run
4 python-3.9 - 26638989809.4 31
77.26
 GitHub Action Run
Source Files on build 26638989809