Translation qualification

Policy oncoref.translation-qualification.2026-11-04.1 assesses 245,535 Ensembl 115 peptide records; 185,993 pass the record-level target rule. Across the 19,893 classified proteins, 19,334 selected representatives qualify. This is annotation-based admission, not experimental proof of production.

Every candidate record receives a classification. The former entry requirement for coding annotation or established ncORF support is now a CTA admission gate; 295 records fail that requirement and remain assessed for normal evidence. Complete inventory audit.

What qualifies

A target record must have an exact source gene/transcript/protein identity, a primary chromosome locus, a coding transcript without LoF or NMD annotation, annotated CDS and complete three-base start and stop codons, no incomplete tags, and an uninterrupted AA20 sequence beginning with methionine. One terminal * is permitted. MANE Select supplies the representative when present on the primary locus; otherwise Ensembl canonical does. Every member must share the same approved sequence. Conflicting designated sequences or unresolved coding evidence require review. A shorter isoform or an alternate locus cannot silently replace a failed representative. Selection method and reviewed exceptions.

Missing mappings, incomplete annotation, NMD, LoF and interrupted sequence create translation review, never normal-tissue Unsafe evidence. Existing Unsafe findings retain precedence. All normal holds and target-gate reasons remain in the audit even when the displayed subtype is Translation or Nested.

These conservative record rules deliberately send selenocysteine-containing sequences (for example SELENOV) to review under the AA20 consumer contract. That is a sequence-support limitation, not a claim that selenoproteins are untranslated. Likewise an absent annotated start codon is a reference-qualification limitation, not proof that no protein can be made.

Worked records

SERPINA2: primary ENST00000553483 / ENSP00000486888 encodes a 309-residue LoF-annotated, incomplete reference record. The 421-residue coding record lies on HSCHR14_7_CTG1 and cannot rescue it. Its current result is Other:Translation.

MROH5: the primary 1,318-residue LoF record is interrupted; two short NMD records remain separately visible. Its current result remains Unsafe through normal evidence. Translation review cannot erase those exclusions.

The worked transcript records preserve source versions, chromosome, locus, biotype, tags, codon/CDS counts and sequence hashes. The full record table is available offline through translation_qualification.records().

Changes and downstream use

Against the pinned evidence PR: 106 classification-label changes, 86 sequence-role changes, and 700 reason changes. Label changes include Other subtype changes; inspect category transitions in the series-wide report. Every changed record and all representative checks are downloadable.

Qualification does not delete the original peptide. The verified sequence bundle retains all records and coordinate-bearing AA20 segments, including exclusions from incomplete or alternate records. Segments never bridge an X, internal stop, selenocysteine or different record. An ncORF nomination is independent evidence; it cannot validate an unrelated NMD reference peptide. This source is the Ensembl peptide inventory and does not claim complete empirical ncORF sequence coverage.

Source: Ensembl release 115, with the exact GTF, peptide, cDNA and ncRNA hashes in the qualification receipt. Reference identity: openvax:reference:v1:sha256:ba6f6186a301a3ef9217ea0199ec3b851f9da2eb43558b7ba7e0d6fcd9cd85e9.

from oncoref import translation_qualification, cta_selection
records = translation_qualification.records()
records[records.symbol.eq("SERPINA2")]
cta_selection.sequence_dispositions()