---
id: DOC-ATLAS-SAMPLE-ORGAN-PROTOCOL-20260906
title: Fixed-panel GSE28866 sample and organ reanalysis
kind: prereg
status: frozen
purpose: Specify descriptive contrasts before inspecting selected peak outcomes.
scope: Exploratory secondary reanalysis of twelve fixed addresses in one 3SEQ matrix.
audience: [maintainers, autonomous research agents]
last_verified: 2026-09-06
---

This is an exploratory secondary reanalysis, not a true prospective preregistration: previous pooled summaries and sensitivity results have been read. This protocol was written after header/first-three-nontarget annotation inspection and before target outcome extraction. CHRNA6 is an external known-marker check. Fixed panel: CHRNA6, CD276, SSTR2, PRAME, FAP, CD248, CSPG4, MSLN, L1CAM, GPC3, ALPP, CDH17. No extra targets or outcome-based peak selection.

Input is the complete GEO GSE28866 normalized 36,048-peak cancer-and-normal gzip, SHA256 11dae64b2d6b6e77846c3f14971fc9a313da86eb52a4b8b83df96c23eedc0ffd. Seven leading annotation columns and 93 library columns are separated explicitly. Preserve all original source fields for every row with an exact case-sensitive symbol token in either gene_symbol or peak_exon_gene_symbol. Split tokens on comma, semicolon, pipe, slash or whitespace, never substring-match. Report both fields, all tokens, multi-symbol assignments and discordance; retain ambiguous rows as qualified, separately labelled estimates, with a strict subset whose union of symbol tokens is exactly the target. No new alias/coordinate remapping: no matched row means unrepresented mapping, not absent expression. Preserve unknown symbols/empty annotations in a full annotation inventory so unassigned peaks remain auditable without assigning them to targets.

Every selected peak and all 93 individual library values are retained. Document every column's STT, class, matched GEO sample records and matching uncertainty. Header-defined non-EMC sarcomas are ESS, EWS, GIST, LMS, MLPS, DDLPS, SS. Other cancers are retained as source values but excluded from primary sarcoma contrasts. Exactly ESS_STT5520_rep1/rep2 and LMS_STT516_rep1/rep2 are averaged arithmetically on the supplied transformed scale at each peak, counting each resulting STT once. Retain both raw values and their range. No other inferred replicate merging. STT labels are analysis units, not proven unique patients.

Per peak, median of four EMC values minus comparator median is the primary signed descriptive contrast. Comparator strata: each of seven sarcoma histologies; pooled 30 non-EMC sarcoma STT units; each of five adult normal organs (17 libraries); pooled adult normals; and separately each fetal organ and all ten fetal libraries. Adult/fetal pooled 27-normal comparison is explicitly secondary to expose mixing. Compute each individual EMC value minus each stratum median, and EMC minimum minus stratum maximum to expose overlap. Record ties, available denominators, and missing/nonfinite cells without imputation; no contrast if either arm empty. Signed direction is exact on decimal source values. Summarize across peaks by counts of positive/zero/negative contrasts and positive/zero/negative individual contrasts; never average unlike peaks or choose a representative peak. Strict and ambiguous mappings remain distinguished.

Descriptive sensitivity: recompute pooled-sarcoma peak contrasts leaving each EMC unit out and leaving each sarcoma histology out; record every result and direction changes including zero. No p-values, clinical thresholds, fitted classifiers, RNA fold changes, or population uncertainty intervals. Values are depth-scaled square-root compressed 3SEQ peak quantities, not absolute abundance. Cancer-selected peaks bias representation; bulk FFPE measurements, limited normal organs and unresolved patient identity/overlap limit inference. PRAME is an intracellular HLA-presented address; this cannot establish surface localization, peptide presentation, safety, efficacy, therapeutic window or independent validation. The earlier feasibility statement that full matrix values were unavailable is superseded only for matrix availability.

Stop after one deterministic computation, independent arithmetic/mapping/replicate checks, source/output hashes and concise decision about whether further atlas work is justified. No manuscript, shared state, commit, publication or preflight. Actual runtime permissions supplied in this agent session: approval_policy=never and sandbox_mode=danger-full-access (developer permissions instructions); no requested elevation. Model family GPT-6 per runtime identity; effort medium per dispatch request (no independent telemetry); usage unknown. Resource paper:PUB-SURFACE-TARGETS:sample-organ; owner /root/atlas_sample_organ; base 95b30a620c45f459463a032a2044b253112b2d4a. 1,800 scientific seconds is a checkpoint, not a paper deadline.
