Versioned exact-protein background policy

For current CTA/oncofetal research decisions, use cta_selection: 546 groups representing 582 loci. This page documents the separately pinned v1.20.0 expression and exact-protein background policy. Its counts and classifications describe that policy only. A candidate selection does not establish that every peptide or isoform is absent from normal tissues.

CTA means a reproductive-associated target candidate that passes this research screen. It does not mean clinically proven safe. The policy documented here is oncoref.target-sets.2026-10-25.2: 213 CTA, 741 Neither and 1,586 Unsafe among 2,540 mapped nominations. Across all 20,097 canonical human genes the counts are 213 / 2,179 / 17,705, respectively.

Versioned report and changes · Four-page PDF brief · All candidate decisions · Reviewed findings and citations · Follow-up questions

The three outcomes

Outcome	Meaning	Target set	Forbidden background
CTA	Affirmative reproductive evidence; no accepted exclusion or recorded material hold; required measurement screen passes	Yes	No
Unsafe	Accepted normal-protein evidence, excessive normal RNA, or an applicable reviewed safety veto	No	Yes
Neither	Unresolved evidence, inadequate target evidence/coverage, or insufficient selectivity	No	No

Unsafe means excluded from unrestricted gene-level targeting, not necessarily observed clinical toxicity. Neither is not a safe reserve list or proof of absence.

Read the rules in this order

1. Identify the molecule and context. Preserve species, reference gene/protein, tissue, cell type, developmental/activation state, assay, units and source version. An unresolved family name does not substitute for a mapped gene. Missing data are unknown, never zero.
2. Exclude credible normal expression. Accepted protein production in a non-permitted normal compartment, curated positive normal IHC (including Low), RNA exceeding its applicable absolute limit, or a reviewed gene-level safety veto gives Unsafe. Any audited non-permitted or unsited normal value >30 TPM or nTPM also gives Unsafe. TPM and nTPM are checked separately.
3. Hold material uncertainty. An unresolved normal-expression/identity conflict, uncertain antibody evidence, a source-specific normal-cell RNA trigger, or a qualified donor-variable finding gives Neither, unless step 2 already supplies a decisive exclusion.
4. Require affirmative reproductive evidence. Admission needs ≥5 nTPM in a reproductive compartment, reliable reproductive protein, or accepted reproductive expression plus tumor expression/antigen recognition evidence. Protein or reviewed biological support can satisfy admission below 5 nTPM. Tumor expression or a trial registration alone cannot.
5. Require quantitative restriction and assessed coverage. Each screened bulk normal tissue must pass both its absolute ceiling and selectivity comparison. The pinned candidate screen, routine RNA panel and identity/quality checks must pass. Failure of admission, coverage or selectivity alone gives Neither.
6. Call the remainder CTA. A source-backed candidate nomination is required; passing a normal-expression screen alone does not make an uncurated gene a CTA.

Every rule is evaluated. The report preserves all applicable reasons and chooses one principal reason for navigation. Reproductive RNA, negative IHC and clinical interest cannot cancel accepted normal-protein evidence. There are no new gene-name branches or therapeutic-familiarity exemptions.

Variable is a narrow reconsideration rule, currently supported for CTAG2 only. Reviewed vital-tissue donors must include reported zero and a failing RNA value in the same tissue/cohort/version/units, with appropriate negative local IHC. It can turn the corresponding ordinary RNA exclusion into Neither; it does not override >30, accepted normal protein, positive curated normal IHC or a safety veto. RNA and IHC volunteers are not assumed to be the same people. An RNA-zero volunteer is not thereby eligible for treatment.

Which tissues count?

The anatomy policy is unchanged. Reproductive and Disposable are permitted; Manageable, Protected and Vital are screened.

Tissue class	Meaning and examples
Reproductive	Union of testicular, other male reproductive, ovarian, female tract, mammary, placental, other extraembryonic and explicitly embryonic/fetal compartments
Disposable	Narrow project allowances: cutaneous melanocytes, hair follicles, tonsil, appendix and gallbladder
Manageable	Whole skin, spleen, thymus, thyroid and adipose; still screened
Protected	Salivary glands, cartilage, lymph nodes and specified mucosal/other organs; still screened
Vital	Critical neurological, sensory and systemic organs, including heart, lung, brain, kidney, liver and adrenal; strictest RNA limits

These are examples, not a replacement for the complete anatomy lists and source mappings. A reproductive organ's resident immune cells are not automatically reproductive. Fetal liver is developmental reproductive context; adult maternal liver remains Vital. Disposable does not supply affirmative reproductive evidence. These are general research sets, not pregnancy- or fertility-preserving clearance.

The tissue rationale review anchors the graded approach in Bradley et al. and organ-specific evidence; the project's assignments and RNA limits are not those papers' validated safety boundaries. Shared anatomy mappings cover human, mouse and dog. This numerical CTA policy accepts human measurements only; it does not transfer human tolerability or thresholds to another species.

Exact RNA and IHC rules

Each cell below gives maximum nTPM / selectivity setting. These are the existing settings, unchanged by this literature update.

Tissue	Missing, uncertain or unusable IHC	Approved negative	Supported negative	Enhanced negative
Vital	2 / 80%	4 / 75%	6 / 70%	12 / 65%
Protected	6 / 75%	8 / 70%	12 / 65%	24 / 60%
Manageable	10 / 70%	20 / 65%	30 / 60%	60 / 55%*

*The independent >30 rule makes the effective Manageable maximum 30, even with Enhanced negative IHC. Equality at an absolute limit passes; >30 fails.

Let R be the highest reproductive bulk nTPM, N the screened normal tissue's nTPM, C its ceiling, and s its selectivity setting as a fraction. Require:

N ≤ C
N ≤ R × (1 − s) / s + 0.1
N < R

The 0.1 nTPM allowance softens low-level ratios; it is not a fitted uncertainty interval. The settings are not literal percentages of whole-body expression. For R = 5 and baseline Vital s = 0.80, the selectivity limit is 1.35 nTPM: N = 1.5 passes the absolute ceiling of 2 but fails selectivity. That isolated selectivity failure leads to Neither, not Unsafe.

Relaxed columns require complete negative routine local IHC, using the weakest usable confidence, plus a reliable reproductive-positive control in the curated gene profile. Incomplete or uncertain staining, or no such positive control, uses baseline limits. This is profile-level support, not proof of the same reagent/batch/donor across tissues or a measured RNA-to-protein conversion. Accepted positive normal staining wins over negative staining elsewhere.

Cell-resolved evidence stays separate

Bulk tissue dilution can hide activated immune cells or rare ocular cells. Their expression is therefore checked with source-specific rules, without mixing native units or using a negative bulk stain to relax unpaired cell RNA.

Source	Trigger for Neither, unless a stronger exclusion applies
HPA single-cell clusters	≥5 nTPM and ≥20 reads in a non-permitted cluster, at least as high as the reproductive-cell maximum; missing reproductive coverage or unresolved anatomy remains a concern
Original DICE distributions	≥5 TPM in ≥3 reported expression values for the same cell type/state
HPA/Monaco immune summaries	≥5 in the source's native TPM or nTPM
Adult corneal atlas	Mean CP10K ≥1, ≥5 positive cells and ≥20 counts in ≥2 annotated sample groups
Primary antibodies	Existing uncertain-profile normal staining, or a specifically reviewed material protein/identity conflict

The >30 TPM/nTPM rule overrides these holds. CP10K is not converted to TPM and retains its separate trigger. DICE distribution columns are not stable donor identities across genes; HPA's DICE summary is not independent replication. Activated nonmalignant cells are normal-cell contexts. Rarity or cytokine-blocking therapy does not automatically permit killing the producing cells. Sources, uniform cache/API and native-unit rules.

What the literature changes

Findings have Accepted, Unresolved or Rejected status. Acceptance is a curation judgement about a specific observation, not a paper-count score. Record identity, endogenous versus engineered expression, normal versus malignant state, anatomy, assay specificity and limitations. Negative results in another context do not refute a positive automatically. Cross-reactivity or an unresolved isoform can prevent a signal from becoming an accepted gene-level finding.

Example	Current decision	Why
FTMT	Unsafe	Normal human neuronal protein with reported specificity controls. Primary study
ABCC11	Unsafe	Reported normal neural protein supported by multiple modalities; abstract-level review, full reagent details still a follow-up. Study
YBX2	Unsafe	Normal vascular/adrenal protein; specimen and polyclonal-control limitations retained. Study
CCDC158	Unsafe	Reported renal tubular protein/function; abstract-level review and isoform follow-up retained. Study
ACTRT1	Unsafe	Endogenous protein in differentiating normal keratinocytes; cultured state is explicit. Study
MAGEC2	Neither / Conflict	Purkinje staining requires reagent/family-specificity resolution. Study
LGALS13	Neither / Conflict	Historical extra-placental reactivity conflicts with placental restriction; identity/anatomy unresolved. Early protein study, placental study
PLCZ1	Neither / Conflict	Qualitative pancreatic NYD-SP27 RNA needs abundance/isoform reconciliation. The paper stains CFTR, not PLCZ1; the prior protein interpretation is rejected. Study
CT55	CTA	Reproductive plus leukemia-recognition evidence satisfies admission below 5 nTPM; normal gates pass. Study
FOXR2	Neither / Antibody	Affirmative reproductive/cancer biology does not resolve normal-antibody concerns. Study
ABCB5, IL5, IL13, PAPPA	Unsafe, unchanged	Previously accepted normal-cell protein production; reproductive enrichment cannot override it. Reviewed source records

Clinical evidence is product-specific. CTAG1A/B and XAGE1A/B remain CTA. PRAME is Unsafe after accepted normal adrenal-protein evidence; MAGEA4 is Neither because its raw normal-antibody signals remain unresolved. Their clinical products retain their own evidence, summarized in the completed adjudication report. A trial, response or lack of observed toxicity for a selected peptide/HLA/receptor does not establish safety of every peptide or proteoform from that gene. Conversely receptor-specific cross-reactivity is not automatically proof of normal expression of the intended target gene.

Review coverage is incomplete. The literature search covered all 2,540 nominations, but only 32 genes received new individual primary-paper review; 46 have specific notes including prior work. The 178 cases remaining from the original 180-gene raw-antibody queue have now all received completed adjudications: 177 Neither and 1 Unsafe; the other two were addressed in the initial eight-gene review. No case from this queue remains CTA. This resolves their classification, not every stain's molecular identity. Other unmeasured contexts and the wider literature remain gaps. See the review coverage and open issues.

The maximum-based selectivity rule can still favor a gene with one high reproductive compartment and diffuse low normal RNA. Continuous normal breadth, empirical assay uncertainty and broad isoform adjudication are not yet selection rules. These limitations remain visible rather than being hidden in an unvalidated weighted score.

API, reproduction and exact proteins

from oncoref import target_sets, cta_policy, normal_rna_review

policy = cta_policy.manifest()      # current policy, thresholds, counts, hashes
findings = cta_policy.findings()   # accepted/unresolved/rejected cited facts
adjudications = cta_policy.adjudications()  # 178 completed antibody decisions
rows = target_sets.results()      # decisions, all reasons, measurement provenance
cta = target_sets.gene_ids("CTA")
neither = target_sets.gene_ids("Neither")
forbidden = normal_rna_review.forbidden_gene_ids()  # includes extra reference loci
followup = normal_rna_review.followup_reviews()

cta_policy.classify(facts, findings) is the pure decision function. It requires explicit, reference-bound measurement-gate facts, not a previous final category. cta_policy.manifest() and the result rows identify the active policy; older classifier module constants identify their historical algorithms.

.venv/bin/python scripts/build_cta_policy.py \
  --output /tmp/cta-policy-report \
  --assets /tmp/cta-policy-proteins \
  --sequence-input /path/to/pinned-original-sequence-assets

The builder checks hashes, reconstructs facts from frozen measurement audits, reproduces the previous categories with the original accepted findings, then applies the reviewed additions. --publish-dataset updates the packaged data after invariants pass. Outputs include all decisions, each reason, changed rows, follow-up, exact proteins and checksums. Historical reports and source data remain unchanged. Current build inventory.

Gene categories and complete exact-sequence protein categories are disjoint. A protein encoded by multiple genes takes Unsafe → Neither → CTA precedence. An unclassified encoding gene is Neither. Gene-level expression does not resolve which isoform is expressed. XAGE1A/B and CTAG1A/B preserve shared protein identity.

The forbidden union is unsafe.fasta.gz plus additional-forbidden.fasta.gz (incomplete/ambiguous translations from Unsafe genes only). Neither contributes to neither target nor default forbidden sets. Full proteins can still share k-mers; subtract the forbidden union without crossing ambiguous residues, stops or record boundaries. Missing complete translations remain coverage gaps.

The 1,721 additional exact reference loci retain their RNA decisions and are reissued under the current policy. The matching v1.20.0 release contains 359 CTA / 3,230 Neither / 126,396 Unsafe complete exact protein sequences. Use matching policy/reference manifests, not an older FASTA with the current list.

Migration: the historical 624-gene CTA API is unchanged. It is a different selection, not an alias for this policy. Downstream software must explicitly adopt target_sets.results() or target_sets.gene_ids("CTA"), record cta_policy.manifest(), and rebuild its peptide exclusion basis from matching release assets. A version upgrade alone does not switch the legacy selection.

The v1.20 sequence-background PDF brief is reproducible with python scripts/build_cta_release_pdf.py using the reports extras. The separately saved scripts/slides/ workflow reproduces the historical 24 October slide deck; it does not relabel it as current.