mutopia 1.0.5

No project description provided

MuTopia learns topographic models of somatic mutation: it simultaneously decomposes a cancer cohort’s mutation counts into distinct processes (signatures) and explains how local genomic context shapes each signature’s activity across the genome.

Unlike standard signature tools that produce per-sample counts, MuTopia produces:

• A genome-wide topographic map of each mutational process at 10 kb resolution

• SHAP feature attributions linking signature activity to chromatin, replication timing, sequence composition, and any other genomic track you supply

• Per-mutation posterior probabilities annotated directly in VCF files

• Accurate refitting for targeted panel and exome sequencing via exposure correction

Features

Feature	Description
G-Tensors	Spatially-indexed data structure integrating genomic feature tracks with mutation calls
Topographic NMF	Gradient-boosted local models + iterative NMF learn signature × context interactions
SHAP interpretability	Built-in SHAP beeswarm, interaction matrix, and track-plot visualizations
VCF annotation	mutopia sbs annotate-vcf assigns every mutation a posterior over components
Panel/exome support	Whitelist BED corrects exposure for non-whole-genome sequencing
Hyperparameter tuning	Optuna-based study management via topo-model study
Genome track plots	Composable genome-browser figures in matplotlib

Installation

MuTopia requires Python 3.11 due to a pinned scikit-learn dependency (1.4.2) used for fast gradient-boosted tree training. We recommend uv — it resolves and installs the full dependency set in seconds and keeps environments reproducible across machines.

With Docker (zero setup)

The pre-built image ships with MuTopia plus all the bioinformatics tools it needs (bedtools, bcftools, tabix, UCSC bigWigAverageOverBed):

docker pull allenlynch/mutopia:latest
docker run --rm -v "$PWD":/workspace allenlynch/mutopia:latest gtensor --help

With uv (recommended for native installs)

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

uv venv --python 3.11 .venv
source .venv/bin/activate
uv pip install mutopia

With conda

conda create -n mutopia -c conda-forge -y python=3.11
conda activate mutopia
pip install mutopia

Verify the CLI tools are on your PATH:

gtensor --help
topo-model --help
mutopia --help

Quick start

# 1. Build a G-Tensor from genomic features and mutation VCFs
gtensor compose config.yaml -w 8

# 2. Split into train / test by chromosome
gtensor split data.nc chr1

# 3. Train a topographic model
topo-model train -ds data.train.nc data.test.nc -k 15 -o model.pkl -@ 8 --lazy

# 4. Analyze results in Python
python - << 'EOF'
import mutopia.analysis as mu

model = mu.load_model("model.pkl")
data  = mu.gt.load_dataset("data.nc", with_samples=False)
data  = model.annot_data(data, threads=8, calc_shap=True)

mu.pl.plot_signature_panel(data)
mu.pl.plot_shap_summary(data, scale=40)
EOF

# 5. Annotate individual sample VCFs
topo-model setup model.pkl data.nc data_setup.nc -@ 8
mutopia sbs annotate-vcf model.pkl data_setup.nc sample.vcf.gz \
    -m mutation-rate.bedgraph.gz \
    -o sample_annotated.vcf

Documentation

Full documentation, tutorials, and API reference are at sigscape.github.io/MuTopia.

Citation

If you use MuTopia in your research, please cite:

Lynch AW, et al. (2025). MuTopia: topographic modeling of mutational processes
across the cancer genome. [Preprint]