Cancer Genomics / TCGA Analysis

TCGA analysis starts with: what cancer type? what data type? Build your cohort FIRST (GDC filters), then analyze. Don't query mutations without defining the cohort — pan-cancer counts from GDCgetmutationfrequency are uninformative without cancer-type context. A mutation frequency of 10% in one cancer type may be 0.5% in another; always specify projectid. Survival analysis (Kaplan-Meier) is hypothesis-generating in retrospective TCGA data — always report sample size and p-value, and note that TCGA cohorts are not treatment-stratified.

LOOK UP DON'T GUESS: never assume TCGA project IDs, NCIt codes, or gene coordinates — use GDClistprojects to confirm project IDs and Progenetixlistfiltering_terms for NCIt codes.

Systematic TCGA/GDC analysis: define cohorts, retrieve clinical data, profile somatic mutations, query copy number variations, run survival analysis, and interpret variants with OncoKB.

When to Use

• "What is the mutation frequency of TP53 in TCGA-BRCA?"
• "Get survival data for TCGA-LUAD patients"
• "Find clinical data for breast cancer cases in GDC"
• "Which TCGA projects have KRAS G12C mutations?"
• "Show CNV amplifications of EGFR in glioblastoma"
• "Annotate BRAF V600E for clinical significance in melanoma"

NOT for (use other skills instead)

• Precision oncology treatment recommendations -> Use tooluniverse-precision-oncology
• Rare disease gene discovery -> Use tooluniverse-rare-disease-genomics
• GWAS variant interpretation -> Use tooluniverse-gwas-snp-interpretation

Workflow Overview

Input (cancer type / gene / TCGA project ID)
  |
  v
Phase 1: Study Selection  -- GDC_list_projects, GDC_search_cases
  |
  v
Phase 2: Clinical Data    -- GDC_get_clinical_data
  |
  v
Phase 3: Somatic Mutations -- GDC_get_ssm_by_gene, GDC_get_mutation_frequency
  |
  v
Phase 4: CNV Analysis     -- Progenetix_cnv_search, Progenetix_search_biosamples
  |
  v
Phase 5: Survival Analysis -- GDC_get_survival
  |
  v
Phase 6: Variant Interpretation -- OncoKB_annotate_variant

Key Identifiers

Common TCGA Project IDs

Phase 1: Study Selection

GDClistprojects: No params required. Returns all GDC/TCGA projects with case counts.

• Use to browse available projects and map cancer types to project IDs.

GDCsearchcases: project_id (string, e.g., "TCGA-BRCA"), size (int, default 10), offset (int). Returns case UUIDs and basic metadata.

• Use to confirm a project exists and retrieve case counts before deeper queries.

Phase 2: Clinical Data

GDCgetclinicaldata: projectid (string), primarysite (string, e.g., "Breast"), diseasetype (string), vitalstatus ("Alive" or "Dead"), gender ("female"/"male"), size (int, 1-100), offset (int). Returns {status, data: [{caseid, demographics: {gender, race, ethnicity, vitalstatus, ageatindex}, diagnoses: [{primarydiagnosis, tumorstage, ageatdiagnosis, daystolastfollowup}], treatments: [{therapeuticagents, treatment_type}]}]}.

• Use project_id + optional filters to retrieve patient-level clinical attributes.
• ageatdiagnosis is in days; divide by 365.25 for years.
• Multiple diagnoses or treatments per case are possible.

# Get clinical data for deceased BRCA patients
result = tu.tools.GDC_get_clinical_data(
    project_id="TCGA-BRCA", vital_status="Dead", size=50
)

Phase 3: Somatic Mutations

GDCgetmutationfrequency: genesymbol (string REQUIRED, alias: gene). Returns pan-cancer SSM occurrence count.

• Returns TOTAL count across all TCGA; no per-project breakdown.
• For cancer-specific data, use GDCgetssmbygene with project_id.

GDCgetssmbygene: genesymbol (string REQUIRED), projectid (string, optional), size (int, 1-100). Returns {status, data: [{ssmid, mutationtype, genomicdnachange, aachange, consequencetype}]}.

• mutation_type: "Single base substitution", "Insertion", "Deletion".
• aa_change: amino acid change notation (e.g., "Val600Glu").

# TP53 mutations in lung adenocarcinoma
mutations = tu.tools.GDC_get_ssm_by_gene(
    gene_symbol="TP53", project_id="TCGA-LUAD", size=50
)

Phase 4: CNV Analysis (Progenetix)

Progenetixsearchbiosamples: filters (string REQUIRED, NCIt code e.g., "NCIT:C4017"), limit (int), skip (int). Returns {status, data: {biosamples: [{biosampleid, histologicaldiagnosis, pathologicalstage, externalreferences}]}}.

• Use to find samples with CNV profiles for a given cancer type.

Progenetixcnvsearch: referencename (string REQUIRED, RefSeq accession), start (int REQUIRED, GRCh38 1-based), end (int REQUIRED), varianttype ("DUP"/"DEL"), filters (string, NCIt code), limit (int). Returns biosamples with CNV in the specified genomic region.

• variant_type="DUP" for amplification, "DEL" for deletion.
• Use filters to restrict to a cancer type.

# EGFR amplifications (chr7:55019017-55211628) in breast cancer
result = tu.tools.Progenetix_cnv_search(
    reference_name="refseq:NC_000007.14",
    start=55019017, end=55211628,
    variant_type="DUP", filters="NCIT:C4017", limit=10
)

Progenetixlistfiltering_terms: No params. Returns all available NCIt codes and labels.

• Use when you need to find the NCIt code for a cancer type.

Progenetixlistcohorts: No params. Returns named cohorts available in Progenetix.

Phase 5: Survival Analysis

GDCgetsurvival: projectid (string REQUIRED, e.g., "TCGA-BRCA"), genesymbol (string, optional -- filters to mutated cases). Returns {status, data: {donors: [{id, time, censored, survivalEstimate}], overallStats: {pValue}}}.

• Each donor has time (days), censored (bool: False=death event, True=censored), and survivalEstimate.
• overallStats.pValue: log-rank p-value (present when gene_symbol splits cohort).
• Without gene_symbol: returns full-cohort survival curve.
• With gene_symbol: returns survival split by mutation status (mutated vs. wild-type).

# Survival for TCGA-BRCA split by TP53 mutation
surv = tu.tools.GDC_get_survival(project_id="TCGA-BRCA", gene_symbol="TP53")
pval = surv["data"]["overallStats"]["pValue"]

Phase 6: Variant Interpretation (OncoKB)

OncoKBannotatevariant: gene (string, alias genesymbol), variant (string, alias alteration, e.g., "V600E"), tumortype (string, OncoTree code e.g., "MEL"). Returns {status, data: {oncogenic, mutationEffect, highestSensitiveLevel, treatments: [{drugs, level, indication}]}}.

• oncogenic: "Oncogenic", "Likely Oncogenic", "Neutral", "Inconclusive", "Unknown".
• highestSensitiveLevel: FDA approval level ("LEVEL1"=FDA-approved, "LEVEL2"=standard of care, etc.).
• Demo mode available for BRAF, TP53, ROS1 without API key.
• Set ONCOKBAPITOKEN for full access.

# Annotate KRAS G12C in lung adenocarcinoma
result = tu.tools.OncoKB_annotate_variant(
    gene="KRAS", variant="G12C", tumor_type="LUAD"
)

Tool Quick Reference

Example Workflows

Workflow 1: Gene-Centric Mutation + Survival Analysis

1. GDC_get_mutation_frequency(gene_symbol="KRAS")
   -> Pan-cancer mutation count

2. GDC_get_ssm_by_gene(gene_symbol="KRAS", project_id="TCGA-LUAD", size=50)
   -> Specific amino acid changes in lung adenocarcinoma

3. GDC_get_survival(project_id="TCGA-LUAD", gene_symbol="KRAS")
   -> Survival split by KRAS mutation status + p-value

4. OncoKB_annotate_variant(gene="KRAS", variant="G12C", tumor_type="LUAD")
   -> Clinical significance + approved therapies (sotorasib)

Workflow 2: Cohort Clinical Summary

1. GDC_list_projects()  -> confirm TCGA-OV exists

2. GDC_get_clinical_data(project_id="TCGA-OV", size=100)
   -> Demographics, tumor stage, treatment history

3. GDC_get_survival(project_id="TCGA-OV")
   -> Baseline overall survival curve for the cohort

Workflow 3: CNV Analysis for a Gene

1. Progenetix_search_biosamples(filters="NCIT:C3058", limit=10)
   -> GBM biosamples with CNV data

2. Progenetix_cnv_search(
       reference_name="refseq:NC_000007.14",
       start=55019017, end=55211628,
       variant_type="DUP", filters="NCIT:C3058"
   )
   -> GBM samples with EGFR amplification

Reasoning Framework

Evidence Grading

Interpretation Guidance

Mutation frequency: A gene mutated in >10% of a TCGA cohort is likely a driver candidate (e.g., TP53 in 36% of all TCGA). Mutations at <1% frequency are typically passengers unless they occur at known hotspots. Always cross-reference with OncoKB oncogenicity annotation.

Survival analysis (Kaplan-Meier): A log-rank p-value < 0.05 suggests the gene mutation is associated with differential survival. Hazard ratio (HR) > 1 indicates worse prognosis for the mutated group. Interpret cautiously: TCGA cohorts are retrospective and not treatment-stratified. Small subgroups (n < 20) produce unreliable survival estimates.

Copy number variation: Focal amplifications (narrow peaks) of oncogenes (EGFR, MYC, ERBB2) are more likely functionally relevant than broad arm-level events. Homozygous deletions of tumor suppressors (CDKN2A, PTEN, RB1) are strong loss-of-function signals. DUP count from Progenetix reflects sample frequency, not copy number magnitude.

Synthesis Questions

A complete cancer genomics report should answer:

1. What are the most frequently mutated genes in this cancer type, and which are known drivers?
2. Does mutation status of the queried gene associate with survival (p < 0.05)?
3. Are recurrent CNV events (amplifications or deletions) present at known oncogene/tumor suppressor loci?
4. What is the OncoKB clinical actionability level for identified variants?
5. How does the mutation landscape compare across TCGA cancer types (pan-cancer context)?

Programmatic Access (Beyond Tools)

When ToolUniverse tools return truncated results or you need bulk data, use the GDC API directly:

import requests, pandas as pd

# Bulk clinical data for a TCGA project
filters = {"op":"and","content":[
    {"op":"=","content":{"field":"project.project_id","value":"TCGA-BRCA"}}
]}
all_cases = []
offset = 0
while True:
    resp = requests.post("https://api.gdc.cancer.gov/cases", json={
        "filters": filters, "size": 500, "from": offset,
        "fields": "submitter_id,demographic.vital_status,demographic.days_to_death,diagnoses.tumor_stage"
    }).json()
    hits = resp["data"]["hits"]
    if not hits: break
    all_cases.extend(hits)
    offset += len(hits)
df = pd.json_normalize(all_cases)

# Download MAF mutation file by UUID
file_uuid = "abc123-..."  # from GDC_list_files result
url = f"https://api.gdc.cancer.gov/data/{file_uuid}"
content = requests.get(url, headers={"Content-Type": "application/json"}).content

# Gene expression: query files endpoint for HTSeq counts
expr_filters = {"op":"and","content":[
    {"op":"=","content":{"field":"cases.project.project_id","value":"TCGA-BRCA"}},
    {"op":"=","content":{"field":"data_type","value":"Gene Expression Quantification"}}
]}

See tooluniverse-data-wrangling skill for pagination, error handling, and format parsing patterns.

Limitations

• GDCgetsurvival with gene_symbol splits on mutation presence only; no multi-gene or stage-based stratification.
• GDCgetmutationfrequency returns pan-cancer total only; per-cancer frequencies require GDCgetssmby_gene per project.
• GDCgetclinical_data returns up to 100 cases per call; use offset for pagination.
• Progenetix uses GRCh38 coordinates; provide GRCh38 positions for Progenetixcnvsearch.
• OncoKBannotatevariant without ONCOKBAPITOKEN operates in demo mode (limited to BRAF, TP53, ROS1).
• Progenetix filters param requires NCIt CURIE format (e.g., "NCIT:C4017"), not free text.