SparseSignatures

SparseSignatures extracts mutational signatures from somatic point mutation data to infer the mutagenic processes that shape cancer genomes.

Key Features:

• User-specified background signature: Incorporates a user-provided background signature into the signature extraction model.
• Regularization: Applies regularization to reduce noise in non-background signatures.
• Cross-validation for model selection: Uses cross-validation to determine the optimal number of signatures.
• Scalability and robustness: Designed to improve robustness, interpretability, and performance on large or heterogeneous datasets.
• Data compatibility: Operates on somatic point mutation data and whole-genome sequencing datasets.
• Benchmark performance: Demonstrated improved accuracy and reliability on simulated datasets across multiple standard evaluation metrics.

Scientific Applications:

• Mutational signature discovery: Identifying mutational signatures—patterns in the rates and spectra of point mutations—to elucidate underlying mutagenic processes.
• Cancer genome analysis: Applied to whole-genome sequences from pancreatic and breast tumors to identify well-differentiated signatures linked to established mutagenic mechanisms and associated with patient clinical characteristics.
• Method benchmarking and validation: Evaluating signature extraction accuracy and reliability using simulated datasets and standard evaluation metrics.

Methodology:

Extracts signatures from somatic point mutation data while incorporating a user-specified background signature, applying regularization to non-background signatures, and using cross-validation to select the optimal number of signatures; performance assessed on simulated datasets.