Tumor Mutational Burden

Institut Curie - TMB analysis

This tool was designed to calculate a Tumor Mutational Burden (TMB) score from a VCF file.

The TMB is usually defined as the total number of non-synonymous mutations per coding area of a tumor genome. This metric is mainly used as a biomarker in clinical practice to determine whether to use or not immunomodulatory anticancer drugs (immune checkpoint inhibitors such as Nivolumab). Whole Exome Sequencing (WES) allows comprehensive measurement of TMB and is considered the gold standard. In practice, as TMB is mainly used in the routine of the clinic, and due to the high cost of WES, TMB calculation based on gene panels is preferred.

Currently, the main limitation of TMB calculation is the lack of standard for its calculation. Therefore, we decided to propose a very versatile tool allowing the user to define exactly which type of variants to use or filter.

---

Tool summary

Installation

Option 1 — conda environment + pip (recommended)

# 1. Create and activate the conda environment
conda env create -f environment.yml
conda activate pyTMB_new

# 2. Install the package (regular)
pip install .

# 3. Or install in editable / development mode
pip install -e .

# 4. To also enable pyEffGenomeSize (requires pybedtools + pandas)
pip install -e ".[effgenomesize]"

After installation two CLI commands are available directly in your $PATH:

pyTMB
pyEffGenomeSize

Option 2 — conda only (pre-built bioconda package)

conda env create -n pytmb
conda activate pytmb
conda install -c bioconda -c conda-forge tmb=1.6.0

Option 3 — run scripts directly (backward-compatible shims)

The bin/ directory still contains thin wrapper scripts that delegate to the installed package. After installing with pip install . you can still call:

python bin/pyTMB.py [options]
python bin/pyEffGenomeSize.py [options]

Recommendations

In order to have homogenous VCF entry files and to avoid VCF ambiguities, we recommend to normalize the VCF files before calculating the TMB. This is especially useful if the VCF file contains Multi Nucleotide Variants (MNVs) or multiallelic variants.

bcftools norm -f FASTA -m -o file_norm.vcf file

Implementation

The idea behind this tool is quite simple. All variants are scanned and filtered according to the criteria provided by the user. If a variant passes all the filters, it is therefore used for the TMB calculation. In other words, if no filters are provided, the tool will simply count the number of variants.

The TMB is defined as the number of variants over the size of the genomic region (in Mb).
To calculate the effective genome size, the user can provide a BED file (--bed) with the design of the assay.
This BED file should be ordered, 0-based and with no header.
Another alternative is to specify the size directly using --effGenomeSize.
Importantly, it is the user’s responsibility to provide the BED corresponding to the VCF input file.

We also provide the pyEffGenomeSize command to calculate the effective size from a BAM file using annotations, coverage and mapping quality thresholds defined by the user.

---

Package structure

Since version 1.6.0, the code is organized as an installable Python package:

pytmb/
├── __init__.py           # version + public API
├── config.py             # loadConfig()
├── vcf_utils.py          # getTag(), getMultiAlleleHeader()
├── genome_size.py        # getEffGenomeSizeFromBed(), getEffGenomeSizeFromMosdepth()
├── filters.py            # isAnnotatedAs(), isPolym(), isCancerHotspot(), …
├── tmb.py                # calculate_tmb() — importable library function
└── cli/
    ├── run_tmb.py        # pyTMB entry point
    └── run_effgenomesize.py  # pyEffGenomeSize entry point

The core calculate_tmb() function can also be used programmatically:

from pytmb import calculate_tmb, loadConfig

db_flags     = loadConfig("config/snpeff.yml")
caller_flags = loadConfig("config/mutect2.yml")

results = calculate_tmb(
    vcf_path="sample.vcf.gz",
    db_flags=db_flags,
    caller_flags=caller_flags,
    eff_genome_size=33_280_000,
    vaf=0.05,
    maf=0.001,
    min_depth=20,
    min_alt_depth=2,
    filter_low_qual=True,
    filter_non_coding=True,
    filter_syn=True,
    filter_polym=True,
    polym_db="1k,gnomad",
)
print("TMB =", results["tmb"])

---

Quick help

pyTMB -h

usage: pyTMB [-h] -i VCF --dbConfig DBCONFIG --varConfig VARCONFIG
             [--sample SAMPLE] [--effGenomeSize EFFGENOMESIZE] [--bed BED]
             [--vaf VAF] [--maf MAF] [--minDepth MINDEPTH]
             [--minAltDepth MINALTDEPTH] [--filterLowQual] [--filterIndels]
             [--filterCoding] [--filterSplice] [--filterNonCoding]
             [--filterSyn] [--filterNonSyn] [--filterCancerHotspot]
             [--filterPolym] [--filterRecurrence] [--polymDb POLYMDB]
             [--cancerDb CANCERDB] [--verbose] [--debug] [--export EXPORT]
             [--version]

Calculate a Tumour Mutational Burden (TMB) score from a VCF file.

options:
  -h, --help            show this help message and exit
  -i VCF, --vcf VCF     Input file (.vcf, .vcf.gz, .bcf, .bcf.gz) (default: None)
  --dbConfig DBCONFIG   Databases config file (YAML) (default: None)
  --varConfig VARCONFIG Variant calling config file (YAML) (default: None)
  --sample SAMPLE       Specify the sample ID to focus on (default: None)
  --effGenomeSize EFFGENOMESIZE
                        Effective genome size (bp) (default: None)
  --bed BED             Capture design BED file (default: None)
  --vaf VAF             Filter variants with Allelic Ratio < vaf (default: 0)
  --maf MAF             Filter variants with MAF > maf (default: 1)
  --minDepth MINDEPTH   Filter variants with depth < minDepth (default: 1)
  --minAltDepth MINALTDEPTH
                        Filter variants with alt-allele depth < minAltDepth (default: 1)
  --filterLowQual       Filter low quality (not PASS) variants (default: False)
  --filterIndels        Filter insertions/deletions (default: False)
  --filterCoding        Filter coding variants (default: False)
  --filterSplice        Filter splice variants (default: False)
  --filterNonCoding     Filter non-coding variants (default: False)
  --filterSyn           Filter synonymous variants (default: False)
  --filterNonSyn        Filter non-synonymous variants (default: False)
  --filterCancerHotspot Filter variants annotated as cancer hotspots (default: False)
  --filterPolym         Filter polymorphism variants (see --maf) (default: False)
  --filterRecurrence    Filter on run-level recurrence values (default: False)
  --polymDb POLYMDB     Databases for polymorphism detection, comma-separated (default: gnomad)
  --cancerDb CANCERDB   Databases for cancer hotspot annotation, comma-separated (default: cosmic)
  --verbose             Activate verbose mode (default: False)
  --debug               Export original VCF with TMB_FILTER tag (default: False)
  --export EXPORT       Export a VCF with passing variants to this path (default: None)
  --version             Version number

---

Configs

Working with VCF files is usually not straightforward, and mainly depends on the variant caller and annotation tools/databases used. In order to make this tool as flexible as possible, we set up two configuration files to define which fields have to be checked and in which case.

The --dbConfig file describes all details about annotation. We provide configurations for:

• Annovar — config/annovar.yml
• snpEff — config/snpeff.yml
• VEP — config/vep.yml

These files can be customized by the user.

The --varConfig file contains all variant-caller-specific parameters. Config files for:

• Varscan2 — config/varscan2.yml
• Mutect2 — config/mutect2.yml
• Strelka — config/strelka.yml

are provided as examples.

The yaml config files list the different key:values for each function. For example, to assess whether a variant is coding (for Annovar):

isCoding:
  Func.refGene:
    - exonic

Regarding databases, the polymorphism fields for Annovar are:

polymDb:
  1k:
    - 1000g2015aug_all
  gnomad:
    - gnomAD_exome_ALL
  esp:
    - esp6500siv2_all
  exac:
    - ExAC_ALL

The user can then choose databases with --polymDb 1k,gnomad,esp,exac.
The same logic applies for --cancerDb.

---

Usage

pyTMB

General parameters

-i

Input file (.vcf, .vcf.gz, .bcf, .bcf.gz)

--sample

Specify the sample ID to focus on. Required when dealing with multi-sample VCFs.

--bed and --effGenomeSize

Specify either a sorted BED file with no header, or the size of the effective genome directly.

Filters

--vaf MINVAF

Filter variants with Allelic Ratio < minVAF. The field name is defined in config/caller.yml. The tool first checks the FORMAT field and then falls back to the INFO field.

--maf MAXMAF

Filter variants with MAF > maxMAF. The databases to use are set with --polymDb and the config/databases.yml file.

--minDepth MINDEPTH

Filter variants with depth < minDepth. The field name is defined in config/caller.yml. The tool first checks the FORMAT field and then falls back to the INFO field.

--minAltDepth MINALTDEPTH

Filter variants with alternative allele depth < minAltDepth. Checked in the FORMAT field.

--filterLowQual

Filter variants for which the FILTER field is not PASS or for which the QUAL value is not null.

--filterIndels

Filter insertion/deletion variants.

--filterCoding

Filter coding variants as defined in the config/databases.yml file.

--filterSplice

Filter splice variants as defined in the config/databases.yml file.

--filterNonCoding

Filter non-coding variants as defined in the config/databases.yml file.

--filterSyn

Filter synonymous variants as defined in the config/databases.yml file.

--filterNonSyn

Filter non-synonymous variants as defined in the config/databases.yml file.

--filterCancerHotspot

Filter variants annotated as cancer hotspots as defined in the config/databases.yml file. All variants with a cancer annotation (e.g. a COSMIC ID) will be removed.

--filterPolym

Filter polymorphism variants from genome databases. The databases can be listed with --polymDb. The fields to scan for each database are defined in config/databases.yml and the population frequency is compared against --maf.

--filterRecurrence

Filter on run-level recurrence values. The VCF must already contain recurrence information as defined in the config/databases.yml file.

Outputs

By default, the tool prints a summary with the calculated TMB value.

--export PATH

Export a VCF file containing only the variants used for TMB calculation.

--debug

Export a VCF file with the tag TMB_FILTERS in the INFO field. This tag contains the reason why each variant would be filtered.

---

pyEffGenomeSize

This tool calculates the effective genome size from a BAM file. This parameter has a strong impact on the TMB result. For instance, if only coding variants are used it makes sense to restrict the denominator to coding regions only.

pyEffGenomeSize -h

usage: pyEffGenomeSize [-h] --bed BED [--gtf GTF] [--bam BAM]
                       [--minCoverage MINCOVERAGE] [--minMapq MINMAPQ]
                       [--filterNonCoding] [--filterCoding]
                       [--featureTypes FEATURETYPES [FEATURETYPES ...]]
                       [--saveIntermediates] [-t THREAD]
                       [--oprefix OPREFIX] [--verbose] [--version]

General parameters

--bed

The input BED file to filter. Should be 0-based, sorted, and with no header. Required.

--gtf

A sorted GTF file for genome annotation (e.g. gencode.v19.annotation.gtf or .gtf.gz).

--bam

A BAM file from your experiment to extract mapping quality and coverage information. When provided, mosdepth is automatically run to filter regions based on coverage and mapping quality.

Filters

--minCoverage

Minimum coverage per region of the BED file. Requires --bam.

--minMapq

Mapping quality threshold. Reads below this value are ignored. Requires --bam.

--filterNonCoding

Remove regions considered non-coding from the GTF/BED intersection to keep only exonic regions. Requires --gtf.

--filterCoding

Remove regions considered coding based on the transcript_type field in the GTF. Requires --gtf and --featureTypes.

--featureTypes

Choose one or more feature types from exon, gene, transcript, UTR, CDS to retain in the final BED file. Default: exon. Required with --filterCoding.

Output / Misc

--saveIntermediates

Keep intermediate files (mosdepth output, filtered GTF, intersect BED) instead of deleting them.

-t, --thread

Number of threads for mosdepth. Default: 1.

--oprefix

Output file prefix. Default: pyeffg.

--verbose

Activate verbose mode.

--version

Show version number.

---

Usage examples and recommendations

Gene Panel

Calculate the TMB on a gene panel VCF (coding size = 1.59 Mb, caller = Varscan2, annotation = Annovar) with the following criteria:

• minDepth at 100×
• non-synonymous, coding and splice variants only
• polymorphisms filtered using 1K, gnomAD databases at MAF 0.001 (0.1%)

pyTMB -i ${VCF} \
    --effGenomeSize 1590000 \
    --dbConfig config/annovar.yml \
    --varConfig config/varscan2.yml \
    --maf 0.001 --minDepth 100 --minAltDepth 2 \
    --filterLowQual \
    --filterNonCoding \
    --filterSplice \
    --filterSyn \
    --filterPolym --polymDb 1k,gnomad \
    --export panel_tmb_variants.vcf.gz \
    > TMB_results.log

Whole Exome Sequencing

For WES, filter low-quality, non-coding, synonymous and polymorphic variants. Indels and splicing variants are kept. An effective genome size of 33 Mb is used.

For Mutect2 + snpEff:

pyTMB -i ${VCF} \
    --effGenomeSize 33280000 \
    --dbConfig config/snpeff.yml \
    --varConfig config/mutect2.yml \
    --vaf 0.05 --maf 0.001 --minDepth 20 --minAltDepth 2 \
    --filterLowQual \
    --filterNonCoding \
    --filterSyn \
    --filterPolym --polymDb 1k,gnomad \
    > TMB_results.log

---

Credits

This pipeline has been written by the bioinformatics core facility in close collaboration with the Clinical Bioinformatics and the Genetics Service of the Institut Curie. Many thanks to the seqOIA-IT team for their help in the development and for the extensive testing of the tool!

If you are using this tool for your own research, please cite:
Dupain, C., Gutman, T., Girard, E. et al. Tumor mutational burden assessment and standardized bioinformatics approach using custom NGS panels in clinical routine. BMC Biol 22, 43 (2024). https://doi.org/10.1186/s12915-024-01839-8

AI Disclosure: Augmented

This project is AI-augmented and utilized AI (e.g., Claude) to:

• Generate boilerplate code and specific utility functions.
• Refactor existing code for better performance and readability.
• Draft unit tests and technical documentation.

Verification: Every AI-generated contribution was manually reviewed, debugged, and integrated into the final codebase.

Contacts

For any question, bug or suggestion, please use the issues system or contact the bioinformatics core facility.