OncoRef 1.22.0: safety and CTA identity

The human research selection contains 546 CTA protein-equivalence groups (582 loci), 18,031 Unsafe, 20 Neither and 1,266 unassigned, across 19,863 groups. MAGE-A4, NY-ESO-1, PRAME, CTAG2 and XAGE1A/B pass the uniform rules.

Full report and funnel · 28-slide PDF · One-page logic PDF · Rules · API

Four independent requirements

  1. Safety: absolute normal RNA limits by tissue tier and usable local IHC, qualifying normal protein and applicable clinical safety findings. Only this assessment can establish Unsafe. Manageable excess blocks CTA under the current policy without automatically establishing Unsafe.
  2. Cancer expression: one measured cancer sample ≥2 TPM or HPA pTPM. There is no tumor-versus-normal ratio and RNA does not prove protein abundance.
  3. Reproductive/developmental association: reproductive ≥5 nTPM or a supported developmental lineage mean ≥5 TPM, with the existing cell-support requirements.
  4. Adult restriction: a qualifying reproductive signal exceeds supported adult normal nTPM, or a qualifying developmental signal exceeds supported adult normal TPM. Vital, Protected and Manageable observations form the adverse comparator. Reproductive and Disposable do not. CP10K remains safety-only.

The last three can block CTA admission but cannot supply Unsafe evidence. Neither requires a measured low-expression or diffuse-Manageable profile. Other incomplete or nonqualifying cases remain unassigned. These are research heuristics; source normalization, sampling and antibody sensitivity limit interpretation.

The 29 October restriction change moves TGIF2-RAB5IF and ATF7-NPFF from the preceding 548-group draft to unassigned. Their developmental means are slightly below their supported normal TPM values. No safety exclusion changed. Individual changes · Comparison receipt.

Source coverage and provenance

This release adds the source-version API and expression-coverage repairs described in the full report. It combines HPA normal panels with direct GTEx v11 and HPA brain coverage, adds Petropoulos preimplantation measurements alongside Tyser gastrulation, and resolves source identities while quarantining conflicting aliases. The saved selection pins its normal HPA measurements to v23; current default RNA loaders can resolve v25.1. Source versions, annotations, processing and units remain distinct fields.

The public audit records four assessment rows per human group, including passes:

from oncoref import cta_selection

rows = cta_selection.results(species="human", category="CTA")
evidence = cta_selection.audit(species="human").explain("PRAME")
assessments = evidence[evidence.layer.str.startswith("Assessment: ")]

Dog's 14 native nominations remain unassigned/unassessed; mouse selection remains unavailable. Human thresholds or clinical evidence never transfer through orthology. Neither a missing observation nor an unavailable species becomes zero expression.

Downstream migration

Pin the package version and selection policy oncoref.cta-selection.2026-10-29. Use cta_selection.results() for this selection and keep its group/member identity and evidence flags. Unassigned has a null category, not Neither or Unsafe. MAGE-A4, PRAME and CTAG2 should not inherit older selection dispositions when adopting this policy.

Historical cta_genes/624-gene compatibility APIs and the v1.20 exact-protein near-self background remain separately versioned. This release does not silently rewrite k-mer backgrounds or product-specific blacklists. A gene-group admission is not clearance of every isoform, peptide, receptor or unmeasured tissue.