pirl-unc / tcrsift / 27231745902

Build # 27231745902

Build Type

push

github

Committed by web-flow

Commit Message

2.72.0: signal-peptide QC + Kozak leaky-scan start selection (#263) (#265)

* 2.72.0: signal-peptide QC + Kozak leaky-scan start selection (#263)

from_contig leader extraction took the ORF prefix before the VDJ with zero
validation, so the longest-ORF rule could begin at an upstream 5'-UTR ATG that
reads in-frame into the VDJ — over-capturing the signal peptide (observed: a
54-aa alpha leader on B1-2 where the real SP is ~19 aa; TRAV16's M0/M13/M35
collapse to one sequence off three upstream AUGs).

Fix: pick the start codon by Kozak context via LEAKY SCANNING, not ORF length.

- _kozak_score (Kozak 1987/1991): -3 purine (+2, +1 more if A) and +4=G (+2);
  0 weak / >=2 adequate / >=4 strong. The bar is 'adequate (-3 purine)', NOT a
  consensus match — the validated TRAV16 start M35 is -3=G/+4=A, adequate but
  not textbook, so requiring consensus would reject real starts.
- _kozak_correct_leader: scan in-frame Mets 5'->3', initiate at the FIRST with
  adequate context (what the ribosome does). No adequate AUG -> fall back to the
  longest-ORF start, flagged weak_kozak (visible, never silent).
- leader_qc: missing/no_met/internal_stop/too_short/too_long/no_h_region/ok over
  a 12-25 aa window + hydrophobic h-region check.
- Extraction records provenance per clone: {chain}_leader_qc, _leader_len,
  _leader_kozak_score, _leader_source (contig_kozak / contig_weak_fallback /
  curated_fallback), _leader_support (n agreeing / n total).
- leader_fallback (CLI --leader-fallback, config assemble.leader_fallback):
  substitute a curated SP (CD8A/CD28/IgK/TRAC/TRBC) for an implausible leader;
  default keeps the extraction + its QC flag.
- validate_sequences surfaces a non-ok leader_qc as an informational note.
- Side fix: from_contig gate now accepts cellranger_dir, not just contigs_dir.

Default behavior is flag-don't-error (per the leaky-scan spec): a leader is
always produced (kozak or weak fallback) and QC-flagged; curated substitution
is opt-in via leader... (continued)

Coverage Stats

10543 of 12626 relevant lines covered (83.5%)

0.84 hits per line

Source File

Press 'n' to go to next uncovered line, 'b' for previous

Source Not Available