OncoRef 1.23.0: reproductive concentration and a common tissue panel

The human selection contains 522 CTA proteins (558 member gene loci), 18,031 Unsafe, 20 Neither and 1,290 unassigned across 19,863 proteins. PRAME, MAGE-A4, NY-ESO-1, XAGE1A/B and CTAG2 remain CTA under the same general rules. These are research classifications, not demonstrations of clinical safety.

Full report and funnel · 30-slide PDF · Diffuseness report PDF · ZNF730 calculation tables · Rules and API

What changed

Outcome	1.22 human selection	1.23.0	Explanation
CTA	546	522	23 lack the required GTEx core; MTCP1 has a diffuse reproductive profile
Unsafe	18,031	18,031	Exact membership unchanged
Neither	20	20	MTCP1 enters; NPHP3-ACAD11 moves to unassigned for missing GTEx coverage
Unassigned	1,266	1,290	24 proteins lack the required GTEx core

The former rule requiring reproductive or developmental expression to exceed the highest normal value is removed. The new questions are independent:

1. Normal-tissue safety: the existing tissue/IHC RNA ceilings and normal-protein rules remain. Only this axis can establish Unsafe. Manageable excess blocks CTA without automatically making a protein Unsafe.
2. Cancer expression: at least one measured cancer sample reaches 2 TPM or HPA pTPM. No cancer-versus-normal ratio is imposed.
3. Reproductive expression: adult/placental HPA RNA reaches 5 nTPM, or a supported embryonic lineage mean reaches 5 TPM with the existing cell-support requirements. Reproductive includes placental, extraembryonic, embryonic and fetal branches.
4. Reproductive concentration: a qualifying adult/placental profile should be much higher than most normal tissues, rather than diffusely similar to them.

Complete GTEx v11 coverage across 52 native contexts for every member locus is required alongside the existing normal bulk, cell and immune panels. Missing measurements remain missing. DICE and corneal coverage is optional; observed signals still count. Coverage losses and unassigned reasons.

How diffuseness works

Use HPA v23 consensus RNA, with 38 normal contexts representing 29 equally weighted anatomical units. For each context compute N / (N + R), where N is normal RNA and R is the largest Reproductive value in that same matrix. Average within an organ, then across organs. Disposable tissues count. Thymus is Permissive and is excluded from both the comparator and the reference; its normal-safety tier is unchanged.

This is an average expression similarity, not the fraction of tissues expressing RNA. If every normal value is tenfold below R, the score is 9.09%; if every normal value equals R, it is 50%. A single extreme organ contributes at most 3.45 percentage points. Every tissue, weight and calculation.

A score above 10% fails the adult/placental route only when R is at least 5 nTPM and all required HPA comparison/reference measurements are present. With complete required normal coverage and no unresolved normal-safety hold, a protein with no remaining qualifying route becomes Neither, never Unsafe. The numeric boundary is a project identity criterion, not a clinical threshold.

This gate alone moves MTCP1 from CTA to Neither: testis 5.1 nTPM, widespread normal values around 1–3 nTPM and a 24.53% score. Its embryonic summary is below the qualifying floor. The same single protein moves at cutoffs from 7.5% through 20%. GTEx is retained as a separate descriptive profile, not a second veto or a source chosen opportunistically for the better score.

Explicit limits

Supported embryonic evidence retains an independent route because the available adult and embryonic panels do not establish a comparable developmental contrast. Of the 523 CTAs entering the diffuseness step, 154 qualify through embryonic expression alone. Seven adult-positive proteins lack a complete HPA contrast: CIMIP3, H2AL1Q, PTTG1IP2, SMIM42, TMEM247, TSPY9 and ZC3H11C. Their required GTEx coverage is complete; their adult concentration remains unassessed, not passed. Both limitations appear in the public evidence flags.

HPA consensus is a cross-dataset summary, not a matched donor experiment. RNA summaries cannot establish antigen presentation, absence from rare cells, or pediatric/pregnancy safety. Protein identities collapse identical selected reference sequences, not every alternative isoform.

The merged release preserves dog managed protein/peptide datasets and the mouse P1A native-evidence seed from 1.22.1–1.22.3. Dog's 14 nominations and mouse's single P1A nomination remain unassigned/unassessed; there is no quantitative native selection policy and no transfer of human safety evidence through orthology.

API and downstream migration

from oncoref import cta_selection

proteins = cta_selection.results(species="human", category="CTA")
reasons = cta_selection.audit(species="human").explain("MTCP1")

Pin package 1.23.0, policy oncoref.cta-selection.2026-10-29.3, and the returned source snapshot. The result schema is unchanged. Preserve member gene identities, evidence flags and null categories. Unassigned is not Neither.

Older cta_genes, exact-protein backgrounds and normal_expression_categories.classify_normal_expression() remain separately versioned products. The last API retains Other → Disposable/Silent under its older RNA/IHC policy; current Neither has no dedicated Disposable-only branch. The migration gap is tracked in #647. This release does not silently rewrite downstream k-mer backgrounds.

The analysis, PDF and slide scripts are committed, with pinned measurements, calculation tables and content receipts. The release was checked with a complete and cached replay producing identical classifications, independent score recalculation, retained-target checks and package/API tests.