# RAS mutations in two public pancreatic cohorts

In the processed cBioPortal releases examined, patients with explicit pancreatic ductal adenocarcinoma (PDAC) histology and mutation-profile membership had reported KRAS mutations in **104/150 (69.3%)** in TCGA PanCancer Atlas and **130/135 (96.3%)** in CPTAC Cell 2021. The descriptive difference is **27.0 percentage points** (CPTAC minus TCGA; 95% interval 18.7–35.0). These are frequencies of reported variants in selected study patients; the comparison does not establish a biological difference in population prevalence.

Data were retrieved on 2026-09-30. The accompanying archive preserves the processed input snapshot, public case lists, TCGA testing matrix, source manifest and hashes, calculation script, row-level audit, result tables, and verification record.

## Patients and denominators

| Selection step | TCGA PanCancer Atlas | CPTAC Cell 2021 |
|---|---:|---:|
| Listed patients / samples | 184 / 184 | 140 / 140 |
| Mutation-profile members, before histology restriction | 179 | 140 |
| Explicit ductal histology | 153 | 135 |
| Explicit ductal histology AND profiled for all three RAS genes | **150** | **135** |
| Explicit ductal histology outside this mutation profile | 3 | 0 |

One sample maps to each patient, so no sample-selection tie-break was necessary. All mapped samples are labeled Primary Solid Tumor. These are counts in the selected public releases, not claims about everyone originally recruited.

**Histology rule.** TCGA primary inclusion requires TUMOR_TYPE = “Pancreas Adenocarcinoma, Ductal Type”; CPTAC requires HISTOLOGY_DIAGNOSIS = “Pancreatic Ductal Adenocarcinoma.” A conflicting explicit non-primary histology would exclude a record; no such conflict occurred among the primary inclusions. TCGA's broad PAAD label alone is insufficient: the cohort also contains 8 records coded 8246/3, 5 coded 8480/3, and 1 coded 8020/3. These were excluded from the primary ductal comparison, along with 15 ambiguous records and 2 records supported by ductal code 8500/3 but lacking the explicit ductal text label. CPTAC's 5 adenosquamous carcinomas were excluded. This is a reproducible clinical-annotation definition, not independent pathology confirmation.

**Testing evidence.** The retrieved sequenced case lists identify 179 TCGA and 140 CPTAC samples. TCGA's gene-panel matrix has mutation WXS entries for exactly those 179 and NA for the other five; its content hash matches the Git LFS pointer. The connector's gene-specific coverage endpoint reports every member of each case list as profiled for each of KRAS, NRAS, and HRAS, with zero not-profiled or unknown members. Since coverage is universal within each explicitly enumerated list, membership supplies the per-sample denominator and the all-three-gene intersection. CPTAC's gene-panel matrix URL returned 404; its testing evidence therefore rests on the complete case list plus universal gene-specific coverage totals, not an assumed panel file.

The script checks list IDs, profile IDs, full counts, mutation membership, and the universal-coverage condition before using this inference. If that condition fails, it stops rather than imputing coverage. The five TCGA nonmembers retain null mutation status. “Not profiled in this release” does not mean never tested elsewhere. TCGA's profile description says 184 exomes, but the explicit membership records support 179; the description is not used as the denominator.

Portal profiling establishes assay inclusion, **not adequate coverage or detection sensitivity at every locus**. No read-level callability evidence was obtained. Missing mutation rows within a profiled sample mean “no reported qualifying mutation,” not proven biological wild type.

## Results

The primary endpoint counts patients with at least one reported nonsynonymous coding or splice-site variant in each gene. All 256 retrieved RAS variant records are labeled Missense_Mutation, so the all-reported-variant sensitivity gives identical primary results. Multiple variants in one patient count once. Copy-number changes, fusions, and non-RAS pathway genes are outside this endpoint.

| Endpoint | TCGA: n/N, % (95% Wilson interval) | CPTAC: n/N, % (95% Wilson interval) |
|---|---|---|
| KRAS | 104/150, 69.3% (61.5–76.2) | 130/135, 96.3% (91.6–98.4) |
| NRAS | 0/150, 0.0% (0.0–2.5) | 0/135, 0.0% (0.0–2.8) |
| HRAS | 1/150, 0.7% (0.1–3.7) | 0/135, 0.0% (0.0–2.8) |
| Any KRAS/NRAS/HRAS | 104/150, 69.3% (61.5–76.2) | 130/135, 96.3% (91.6–98.4) |

The single HRAS call is A134T in TCGA-IB-7651-01, which also carries a KRAS call; therefore any-RAS is not the sum of gene counts. Its clinical significance was not adjudicated. Mutation-status fields contain missing markers (“.” or “NA”); these results use the study mutation profiles and do not independently establish somatic origin or oncogenicity for every record. Zero NRAS calls does not establish zero true prevalence.

Intervals describe binomial uncertainty conditional on the selected data. Difference intervals use the Newcombe method based on independent Wilson intervals. They assume independent cohorts, are not multiplicity-adjusted, and omit systematic uncertainty from selection, annotation, tumor content, and detection.

## Can these cohorts be fairly compared?

| Characteristic among primary eligible, profiled patients | TCGA (150) | CPTAC (135) |
|---|---:|---:|
| Median age, years (IQR) | 65.5 (57–73) | 65 (60–71) |
| Female | 70 (46.7%) | 64 (47.4%) |
| Recorded stage I / II / III / IV / missing | 12 / 129 / 4 / 4 / 1 | 23 / 56 / 41 / 9 / 6 |
| Staging edition available | 150: 5th/6th/7th editions | 0 |
| Numeric neoplastic cellularity available | 0 | 76 (59 missing/non-numeric) |
| Neoadjuvant-treatment history available | 150, all “No” | 0 |
| Primary Solid Tumor label | 150 | 135 |

Age and sex are similar, but the recorded stage distributions differ markedly. Broad stages were harmonized as labels only; TCGA uses multiple AJCC editions and CPTAC lacks an edition field in the retrieved catalog. Stage groups cannot be treated as fully equivalent. Both cohorts supply primary tumors, but this does not establish equivalent recruitment, tissue handling, treatment exposure, or surgical eligibility.

TCGA's mutation profile is described as whole-exome sequencing; CPTAC's study description reports WGS and WES. Both study metadata records list hg19, but common reference-genome labels do not imply common capture, depth, calling pipeline, filters, or sensitivity. Tumor cellularity could affect detection, but its contribution to the observed gap cannot be quantified here: TCGA has no comparable cellularity field and CPTAC's field is incomplete. Clinical tables also do not support treatment-matched comparison.

The chosen studies are separate named TCGA and CPTAC releases, rather than two releases of the same TCGA cohort. No patient IDs match directly, but the identifier namespaces differ; this is not definitive person-level exclusion of overlap. Recruitment and calling methods were not independently reconstructed from full publication methods. The preserved sources support the stated metadata, not a claim of complete methodological equivalence.

A descriptive comparison is justified. An adjusted, causal, or population-representative interpretation is not justified by these tables.

## Sensitivity analyses

| Definition | TCGA KRAS | CPTAC KRAS |
|---|---|---|
| Primary explicit ductal label | 104/150 (69.3%) | 130/135 (96.3%) |
| Ductal text label OR compatible ductal code | 105/152 (69.1%) | 130/135 (96.3%) |
| Broader adenocarcinoma annotations, excluding specified non-primary histologies | 115/166 (69.3%) | 130/135 (96.3%) |
| Whole study, without histology filtering | 117/179 (65.4%) | 135/140 (96.4%) |
| Primary definition, recorded stages I–II only | 99/141 (70.2%) | 78/79 (98.7%) |

The gap persists under these definitions. The stages I–II comparison is exploratory, because staging editions are not aligned; it is not an adjusted estimate. Restricting to primary tumors changes nothing because all sample records already carry that label. Matched-cellularity and matched-treatment analyses are unavailable. Whole-study results reproduce the connector's aggregates but are not PDAC-only estimates.

## Reproducibility and checks

Extract the archive and run:

```text
python analyze_ras.py ras_connector_snapshot.json --out rerun_results
```

The script needs only Python's standard library, performs no network calls, and refuses to overwrite an existing output directory. Python 3.12.13 was used here. An independent CLI-path rerun reproduced all nine generated result/audit files byte-for-byte. Separate checks recomputed primary counts from source clinical labels, confirmed one sample per patient, reconciled coverage partitions and mutation membership, verified the TCGA matrix hash, and demonstrated rejection of inconsistent coverage. No row-cap truncation or exact duplicate mutation records were found.

The snapshot is about 4.18 MB uncompressed. The source manifest preserves retrieval timestamps, exact connector arguments, URLs, and SHA-256 checksums. GitHub URLs use the mutable master branch; frozen contents and hashes, rather than future live responses, define this analysis. Scientific limits listed above remain unresolved by computational verification.

Sources: [TCGA cBioPortal study](https://www.cbioportal.org/study/summary?id=paad_tcga_pan_can_atlas_2018); [CPTAC cBioPortal study](https://www.cbioportal.org/study/summary?id=paad_cptac_2021); [cBioPortal DataHub](https://github.com/cBioPortal/datahub); [CPTAC publication record cited by study metadata, PMID 34534465](https://pubmed.ncbi.nlm.nih.gov/34534465/). Exact metadata and case-list URLs are in source_manifest.json.
