MSA

MSA attributes mutational signatures to individual samples using Non-Negative Least Squares and simulation-based optimization to improve assignment accuracy and quantify uncertainty.

Key Features:

• Single-sample attribution: Performs signature attribution on a per-sample basis, addressing noisy data and closely resembling signatures.
• Non-Negative Least Squares (NNLS): Uses NNLS with configurable simulation-based optimization for signature assignment.
• Parametric bootstrap: Applies parametric bootstrap to estimate statistical uncertainties and produce confidence intervals for attributions.
• Mutation class support: Supports single and doublet base substitutions (SBS, DBS), insertions and deletions (indels), and structural variants (SVs).
• Validation framework: Validated using simulations with reference COSMIC signatures and randomly generated signatures.
• Implementation: Implemented as Python scripts orchestrated within a Nextflow pipeline to enable reproducibility and scalability in high-performance computing environments.
• Containerization: Packaged via containers for reproducible deployment.

Scientific Applications:

• Mutational process inference: Assigns known mutational signatures to samples to infer contributing mutagenic processes.
• Uncertainty quantification: Provides confidence intervals for signature attributions to quantify reliability under noisy conditions.
• Benchmarking and validation: Enables benchmarking of attribution performance using COSMIC reference signatures and synthetic signatures.
• Cross-class mutation analysis: Evaluates signature contributions across SBS, DBS, indels, and SV-associated mutation catalogs.

Methodology:

Uses Non-Negative Least Squares with configurable simulation-based optimization and parametric bootstrap for uncertainty estimation; validation via simulations with reference COSMIC signatures and randomly generated signatures; implemented as Python scripts within a Nextflow pipeline and packaged in containers; Non-Negative Matrix Factorisation is noted as the typical extraction method for mutational signatures.