signeR

signeR estimates mutational signatures from single nucleotide variation (SNV) counts in cancer genomes using an empirical Bayesian enhancement of non-negative matrix factorization (NMF).

Key Features:

• Empirical Bayesian Framework: Employs an empirical Bayesian treatment of the NMF model to estimate mutational signatures from SNV count matrices.
• Model Selection: Treats determination of the number of mutational signatures as a model selection problem.
• Robustness to Initial Conditions: Reduces sensitivity to NMF initialization and to issues arising from nonconvexity and high dimensionality of SNV data.
• Clinical Relevance: Introduces two novel concepts for evaluating mutational profiles to aid interpretation of tumor etiologies.

Scientific Applications:

• Cancer Research: Identifies mutational signatures to help characterize cancer origins and group tumors with shared etiological processes.
• Data Analysis: Applicable to analysis of real and synthetic SNV count datasets for benchmarking and comparison with other methods.

Methodology:

Uses an empirical Bayesian treatment of the NMF model to statistically estimate mutational signatures from SNV counts and treats the number of signatures as a model selection problem.