← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-clinical-databases-tumor-mutational-burden / references / usage-guide.md

Tumor Mutational Burden - Usage Guide

Overview

Calculate tumor mutational burden from somatic VCFs with Friends of Cancer Research (Vega 2021) harmonization equations for per-assay calibration to the FDA pembrolizumab 10 mut/Mb pan-tumor threshold (TSO500 = 7.8, Oncomine = 8.4 per Ramos-Paradas 2021). Covers FoundationOne CDx convention of including synonymous variants, hypermutator / ultra-hypermutator tiering (POLE+MMR), tumor-type-specific cutoffs per McGrail 2021 (poor TMB performance in breast / prostate / glioma), blood TMB caveats (BFAST Cohort C failed), and integration with MSI-H, HLA-LOH (LOHHLA / DASH), and neoantigen quality (Luksza fitness, McGranahan clonality).

Prerequisites

pip install cyvcf2 pandas numpy

# VEP for annotation (or snpEff / Funcotator)
# Ensembl VEP: docker pull ensemblorg/ensembl-vep
# Or: conda install -c bioconda ensembl-vep

# HLA-LOH
# LOHHLA: https://github.com/mskcc/lohhla
# DASH: https://github.com/RamratnamLab/DASH

# Friends of Cancer Research TMB harmonization tool
# https://friendsofcancerresearch.org/tmb/

Quick Start

Tell the agent what to do: - "Calculate TMB from this FoundationOne-style somatic VCF using 0.8 Mb scored region (NOT 1.1 Mb panel total)" - "Apply cross-panel calibration: report TMB-H equivalent thresholds for TSO500 (7.8), Oncomine (8.4) per Ramos-Paradas 2021" - "For this MSK-IMPACT cohort, compute TMB excluding synonymous (vs FoundationOne which includes them); document convention" - "Integrate TMB + MSI-H + HLA-LOH (LOHHLA) for ICI eligibility decision" - "For breast / prostate / glioma cohorts, apply tumor-type-specific Samstein 2019 cutoff (NOT universal 10/Mb; McGrail 2021)"

Example Prompts

Standard TMB Calculation

"Compute TMB from this VEP-annotated Mutect2 VCF. Use FoundationOne CDx conventions: scored region 0.8 Mb, include synonymous, VAF >= 5%, exclude hotspots, gnomAD grpmax FAF95 < 0.5%."

"For my TSO500 panel cohort, compute TMB with 1.3 Mb scored region; apply the TSO500 equivalent threshold of 7.8/Mb (Ramos-Paradas 2021) for TMB-H classification."

Cross-Panel Comparison

"Reconcile TMB values across FoundationOne CDx (0.8 Mb scored, includes synonymous), TSO500 (1.3 Mb scored), MSK-IMPACT v4 (1.22 Mb scored). Document counting conventions."

"Audit these published TMB values: are they comparable across studies using different panels?"

Cancer-Type-Specific Cutoffs

"For this breast cancer cohort, do NOT apply universal 10/Mb cutoff. Use tumor-type-specific cutoff per Samstein 2019 (top 20% within tumor type). Cite McGrail 2021 for the breast / prostate / glioma limitation."

MSI + HLA-LOH Integration

"For each tumor: compute TMB, run MSI testing, run LOHHLA. Report ICI eligibility with: MSI-H supersedes TMB-H; TMB-H + HLA-LOH-positive has reduced neoantigen presentation."

Blood TMB

"Compute bTMB from cfDNA panel; flag low tumor fraction (BFAST Cohort C failure mechanism); recommend tissue TMB confirmation."

Hypermutator Characterization

"For samples with TMB > 100 mut/Mb, identify mechanism: MMR-D (SBS6/15/26/44 + ID1/2) vs POLE-exo (SBS10a/10b). For TMB > 500 mut/Mb, suspect POLE+MMR concurrent."

What the Agent Will Do

  1. Verify VCF has VEP / snpEff / Funcotator consequence annotations.
  2. Apply filters: PASS-only, VAF >= 5% (FoundationOne) or higher, depth >= 100, tumor-only germline filter gnomAD grpmax FAF95 <= 0.5%, exclude COSMIC hotspots.
  3. Count nonsynonymous coding variants (configurable to include synonymous for FoundationOne CDx compatibility).
  4. Divide by SCORED region in Mb (NOT total panel size).
  5. Apply Vega 2021 calibration: assay-specific equivalent threshold for FoundationOne 10/Mb sensitivity.
  6. Classify: TMB-low; TMB-H (>= cutoff); hypermutator (>= 100); ultra-hypermutator (>= 500).
  7. For breast / prostate / glioma, apply tumor-type-specific Samstein 2019 cutoff and document the McGrail 2021 tumor-type limitation.
  8. Integrate MSI status: MSI-H is the primary biomarker (Sha 2020); TMB-H is not additive.
  9. Run LOHHLA / DASH for HLA-LOH; flag tumors with reduced neoantigen presentation.
  10. For research-grade neoantigen quality, compute Luksza fitness + McGranahan clonality.

Tips