Recalibrated CTA research selection
215 CTA research candidates under oncoref.cta-research-selection.2026-10-26.1.
Complete list · All nominations ·
Every change · Reviewed clinical comparisons.
The user requested a common evidence policy calibrated to include PRAME and MAGE-A4 alongside genes of comparable quality. Both already pass the numerical RNA gates. This revision changes the evidence threshold for research selection, not the RNA measurements or cutoffs. NY-ESO-1 remains represented by CTAG1A/B. This is benchmark-driven calibration, not independent prospective validation.
| Gene | Earlier expression category | Research selection |
|---|---|---|
| MAGEA4 | Neither | CTA |
| PRAME | Unsafe | CTA |
Common evidence rule
The previous expression-qualified candidates remain selected. A new clinical qualification route requires a prospective treated cohort, repeated objective tumor responses (more than one; stable disease and isolated lesion shrinkage do not count), target-specific evidence, and review of normal selectivity. Every active normal finding must be linked and considered; adding a new finding invalidates the review until reconsidered. An unresolved target-mediated organ toxicity concern or substantial accepted normal protein prevents this route. An unresolved target-mediated organ toxicity concern also holds an otherwise expression-qualified candidate for review.
All measured RNA/IHC gates, reproductive qualification, coverage, hard >30 RNA exclusions, donor variability and cell-RNA holds still apply. Neither a familiar gene symbol nor trial registration qualifies a candidate. No breadth cutoff is used. These criteria are a research evidence threshold, not a clinical safety score or a claim that all encoded peptides can be targeted.
PRAME retains the accepted low adrenal-protein observation. Its selected clinical product has prospective responses and tested peptide selectivity; these support continued target development without asserting that adrenal protein is absent. MAGE-A4 retains the unresolved HPA antibody findings; independent assay-validation evidence and clinical responses support research selection without declaring those stains experimentally resolved.
MAGE-A1 and MAGE-A10 were assessed as comparisons. MAGE-A1's IMA202 cohort had no RECIST objective response; MAGE-A10's NSCLC study reported one response after repeat treatment with subsequent severe complications. Neither establishes the repeated-response criterion. MAGE-A3/12 remain subject to their existing hard RNA/safety exclusions. All findings and qualification limits are exported.
API and sequence scope
from oncoref import cta_research_policy
genes = cta_research_policy.gene_ids()
decisions = cta_research_policy.results()
This is the new research selection. target_sets and cta_policy preserve
the released expression-restriction policy and its exact protein/background
assets. The new table explicitly retains Expression category and Expression
forbidden; it does not export a misleading new whole-protein forbidden set.
Clinical evidence for one peptide/receptor cannot remove every peptide of that
protein from a normal-expression background. Downstream target development
must carry the evidence scope and adjudicate the intended peptide/product.
The near-self policy boundary is tracked in #640; the verified sequence, occurrence and coverage-gap loader is tracked in #641. This research selection does not implement that loader or validate HLA/similarity thresholds. Consumers must not subtract the new candidate list from the forbidden reference, reapply an older Medium/High-IHC filter, or treat Neither as absent normal expression.
Literature work (#638)
Per-gene review status and remaining work covers all 2,540 mapped nominations. Search leads, prior specific assessments, completed antibody dispositions and the new clinical comparisons remain distinct. Unmapped identity reviews retain BAGE and GAGE2B outside that denominator, without substituting a paralog or inferring absence. #638 is not complete. The per-gene table does not substitute for reading the remaining primary papers or expanding truncated searches. No-hit searches are not evidence of absence. The initial scripts/tables were already published in PR #639; the older issue's statement that they remained local is stale.
The new retrieval mode supports cursor pagination with per-page byte receipts and gene-context variants for ambiguous IMPACT, SPATIAL, FATE and MRP8 terms. The original ledger reported 2,491,175 hits across 1,688 truncated queries; these counts include substantial name ambiguity. The selected expansion covers 27 queries for 10 clinical/ambiguity cases; 27 have complete pagination of those query variants. Search ledger and paper leads remain retrieval evidence, not paper adjudication. The variant searches can lose ambiguous aliases and do not claim systematic completeness; the original queries remain available for sensitivity review. Primary-paper receipts record the retrieved versions used for the targeted clinical review.
The breadth report supplies descriptive profiles and missing-coverage bounds without changing memberships. Fetal and embryonic targets need the separate developmental-source assessment.
Reproduction
Run .venv/bin/python scripts/build_cta_research_policy.py --out <empty-directory>
--expansion docs/audits/cta-research-20261026
--paper-receipts docs/audits/cta-research-20261026/paper-receipts.json.
--publish-dataset copies the reviewed research selection to the package;
historical expression decisions and immutable reports remain unchanged.