Verified CTA sequence records

For current CTA transcript and protein accessions without downloading the bundle, use oncoref.cta_representatives.records("BOLL", annotation_release=115). The result includes sequence length, SHA-256 and the pinned reference. Use the bundle for amino-acid bytes and all source occurrences. The legacy Canonical_Transcript_ID annotation is not the current selection contract.

The current sequence bundle binds source proteins and every occurrence to the selection, translation and consumer-role policies. It preserves original sequence bytes, ambiguous residues, incomplete records and explicit coverage gaps. The bundle audit gives its exact identity, counts and comparison with the historical v1.20.0 forbidden assets.

Acquisition is explicit and bounded by the packaged asset hash and size. Reading the descriptor or loading an existing bundle performs no network access.

from oncoref import sequence_bundle

pin = sequence_bundle.descriptor()
path = sequence_bundle.ensure(download=True)  # explicit one-asset acquisition
bundle = sequence_bundle.load(path)          # verifies current package dependencies
records = bundle.records()                  # original bytes and all occurrences
segments = bundle.segments(role="exclude", min_length=8)
coverage = bundle.coverage

ONCOREF_SEQUENCE_CACHE_DIR or ensure(cache_dir=...) selects an isolated cache. A missing or corrupt cache raises an error; an explicit download=True permits repair using the pinned asset. Mixed policy/reference bundles fail validation. For historical use, supply a separately trusted expected_id and require_current=False; an identity read from an untrusted bundle is not a pin.

Each segment has zero-based, half-open coordinates in its original sequence. Segments never bridge an X, internal stop, selenocysteine or separate record. The current AA20 contract requires review for unsupported residues; that is not a claim of absent biological production. Original sequences remain available.

groups() groups exact original sequence strings and retains all records and occurrences. Their combined role follows exclude > review > allow. A consumer matching peptides must likewise aggregate every matching segment occurrence; an allowed CTA occurrence never cancels another occurrence's exclusion or review. Exact sequence grouping is not a validated approximate similarity rule.

An Unsafe gene conservatively excludes its source isoforms, while positive allowance requires the qualified selected representative. Other isoforms require review. Gene-level evidence does not demonstrate isoform-specific production. Missing translations, source genes without decisions, the empty legacy outside-inventory list and ncORF nominations without supplied peptide bytes remain explicit in coverage.

The optional oncoref[quantitative] extra supports bundle.as_orfdata(). It loads the preserved native ORFdata proteome and verifies its original content identity. The returned raw proteome does not apply CTA policy roles; retain the bundle when performing exclusion/review decisions. The current source is the full Ensembl 115 peptide inventory, not a complete empirical ncORF sequence database.

Historical Unsafe plus additional-forbidden FASTAs retain their original policy and reference identities. The downloadable transition table compares exact full sequence hashes; an old sequence absent from the new source requires coverage review rather than implicit allowance. Neither product establishes patient-HLA presentation, receptor recognition or clinical safety.