Thyroid Cancer Multi-Omics

Reproducible computational analysis of CARIS thyroid cancer specimens integrating somatic DNA drivers, RNA-seq disease-status modeling, inflammasome biology, demographic associations, and staged DNA-RNA integration.

Scientific questions

This project asks whether molecular features of thyroid cancer differ by clinical disease status and demographic factors, with emphasis on:

• somatic driver architecture and canonical MAPK groups;
• AWD (alive with disease) versus NED (no evidence of disease) RNA expression;
• inflammasome and thyroid-relevant transcriptional programs;
• race-associated differences in driver composition;
• TERT promoter alterations;
• future integration of CNVs, fusions, autoimmune-thyroiditis annotations, and TCGA-THCA validation.

Analysis architecture

Clinical metadata harmonization
        |
        +-----------------------------+
        |                             |
      RNA-seq                       Somatic DNA
        |                             |
Salmon counts                     QC + MAF processing
        |                             |
filtering / DESeq2                driver reconciliation
        |                             |
VST / PCA                        patient-level driver matrix
        |                             |
AWD vs NED DE                    BRAF / RAS / TERT / MAPK groups
        |                             |
race-adjusted sensitivity        demographic + outcome testing
        |                             |
thyroid programs                 mutation-only pathway summaries
        |                             |
inflammasome models              TERT hotspot characterization
        |                             |
        +-------------+---------------+
                      |
             staged DNA-RNA integration
                      |
          CNV / fusion / thyroiditis /
             TCGA-THCA validation

Current cohort

RNA layer

• 130 final non-medullary AWD/NED samples
• 91 NED
• 39 AWD
• 24,364 genes tested
• 107 samples with reported race for race-adjusted sensitivity analysis

DNA layer

• 139 somatic MAF samples profiled
• 121 with clinical metadata
• 101 in the AWD/NED clinical subset
• 68 AWD
• 33 NED

The DNA landscape is currently strongest for mutation-based analyses. Final confirmatory disparity/outcome analyses should use a consistently harmonized non-medullary cohort and prespecified histology grouping.

RNA-seq findings

The primary AWD versus NED DESeq2 model identified:

• 354 genes at FDR < 0.05
• 110 genes higher in AWD with |log2FC| >= 1
• 41 genes higher in NED with |log2FC| >= 1

Leading AWD-associated genes include IL1A, GJB2, GJB6, SLC2A1, LAMA3, KRT15, PKP1, DSC3, and IL1B.

A race-adjusted sensitivity analysis in 107 samples retained 230 FDR-significant genes, supporting persistence of a substantial disease-status transcriptional signal after adjustment for one major demographic covariate.

Inflammasome findings

At the individual-gene level, CASP4 is higher in AWD:

• median AWD = 8.58
• median NED = 7.95
• nominal P = 0.015
• FDR = 0.20

The composite inflammasome score is not significant in the simple unadjusted AWD/NED comparison, but in a complete-case multivariable model AWD is associated with a higher score after adjustment for race, age, sex, and histology:

• beta = 0.367
• P = 0.0317

Race-associated differences are also visible, including a lower adjusted composite score among Black/African American versus White patients:

• beta = -0.439
• P = 0.0285

These results remain discovery-level and require robust sensitivity analysis and verified autoimmune-thyroiditis annotation.

Somatic driver landscape

The 139-sample mutation cohort is MAPK-dominant.

Primary driver prevalence

• BRAF: 68/139 (48.9%)
• HRAS: 10/139 (7.2%)
• NRAS: 9/139 (6.5%)
• PIK3CA: 5/139 (3.6%)
• PTEN: 5/139 (3.6%)
• KRAS: 4/139 (2.9%)
• RET: 4/139 (2.9%)
• TP53: 4/139 (2.9%)

Canonical MAPK groups

• BRAF V600E: 64/139 (46.0%)
• RAS hotspot: 19/139 (13.7%)
• Other / no canonical MAPK hotspot: 56/139 (40.3%)

Pathway-level mutation prevalence:

• MAPK driver alteration: 69.1%
• PI3K/AKT/mTOR: 8.6%
• TP53/RB: 5.0%
• WNT: 2.2%
• DNA damage response: 1.4%

Race-associated driver architecture

In the current exploratory Black versus White comparison:

• canonical MAPK group distribution differs by Fisher exact test, P = 0.000583;
• RAS hotspot: 39.1% vs 4.9%;
• BRAF V600E: 26.1% vs 50.8%;
• NRAS: 30.4% vs 1.6%, gene-level FDR = 0.0035.

An adjusted NRAS sensitivity model remains significant after histology, age, and sex adjustment:

• OR approximately 18.9
• P = 0.00071

These results are preliminary and require final harmonized cohort filtering and robustness checks before strong disparity claims.

Disease-status mutation results

Mutation-only AWD/NED differences are comparatively weak.

• BRAF: 48.6% AWD vs 50.0% NED
• NRAS: 11.1% AWD vs 0% NED, nominal P = 0.049 but not FDR-significant
• MAPK pathway: 70.8% AWD vs 65.8% NED
• PI3K/AKT/mTOR: 8.3% AWD vs 7.9% NED

No tested mutation-only pathway is FDR-significant.

This is why CNV and fusion integration remains necessary before drawing pathway-level prognostic conclusions.

TERT promoter alterations

Canonical TERT promoter mutations were detected in:

• 27/139 tumors (19.4%)
• C228T: 23/27 TERT-mutant tumors
• C250T: 4/27 TERT-mutant tumors
• median VAF = 0.351
• median alternate reads = 33

TERT is retained as an explicit molecular feature for future outcome and DNA-RNA analyses.

Current completion boundary

Completed / substantially advanced

• RNA AWD/NED differential expression
• race-adjusted RNA sensitivity
• thyroid-program scoring
• inflammasome AWD/NED and demographic modeling
• somatic mutation QC and driver reconciliation
• patient-level driver matrix
• canonical MAPK molecular groups
• race-associated mutation analyses
• mutation-only AWD/NED comparisons
• TERT hotspot characterization

Not yet complete

• finalized CNV analysis
• actionable fusion analysis
• COSMIC mutational-signature analysis
• verified autoimmune-thyroiditis immune deconvolution
• completed matched DNA-RNA integration
• harmonized CARIS-TCGA validation

These unfinished layers are documented as a roadmap and are not presented as completed results.

Repository structure

.
├── README.md
├── config/
├── data/
│   ├── README.md
│   └── synthetic/
├── docs/
├── scripts/
│   ├── R/
│   └── python/
└── results/
    ├── README.md
    ├── figures/
    └── tables/

Active R workflow

1. 00_setup.R
2. 01_build_analysis_metadata.R
3. 02_prepare_rna_counts.R
4. 03_deseq2_awd_ned.R
5. 04_race_adjusted_rna_sensitivity.R
6. 05_thyroid_programs_inflammasome.R
7. 06_driver_reconciliation.R
8. 07_mapk_tert_groups.R
9. 08_dna_clinical_associations.R
10. 09_matched_dna_rna_integration_scaffold.R
11. 10_future_layers_cnv_fusions_tcga.R
12. 11_make_portfolio_outputs.R

Selected portfolio figures

RNA differential-expression summary

Primary driver prevalence

Canonical MAPK molecular groups

Race-associated MAPK architecture

Inflammasome summary

TERT promoter landscape

Privacy

Restricted patient-level clinical, sequencing, and molecular data are not distributed here. Public code is designed around synthetic, aggregate, public, or otherwise approved inputs.

License and reuse

This repository is shared for research transparency and portfolio demonstration. No open-source license is currently granted for reuse, redistribution, or derivative works. Please contact the author regarding reuse of code or materials.