ROBUST DISCOVERY OF MUTATIONAL SIGNATURES USING POWER POSTERIORS

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Mutational signatures reveal molecular mechanisms of carcinogenesis and DNA repair.
• Non-negative matrix factorization (NMF) is a common tool for signature discovery but can be sensitive to model inaccuracies.
• Improved methods are needed for accurate and reliable mutational signature analysis.

Purpose of the Study:

• To develop a more robust and accurate method for mutational signature analysis.
• To improve the identification of cancer-related mutational signatures using a Bayesian NMF model.
• To automatically infer the number of active mutational signatures.

Main Methods:

• A fully Bayesian Non-negative Matrix Factorization (NMF) model utilizing a power posterior.
• Incorporation of a sparsity-inducing prior for automatic signature number inference.
• Validation through extensive simulation studies and analysis of Pan-Cancer whole-genome sequencing data.

Main Results:

• The proposed Bayesian NMF approach demonstrates superior accuracy in recovering true mutational signatures compared to leading methods.
• The method shows improved robustness against model misspecification.
• Accurate recovery of more mutational signatures was achieved on real-world cancer genomics data.

Conclusions:

• The developed Bayesian NMF method offers a more reliable and accurate approach for mutational signature analysis.
• This advancement can enhance our understanding of cancer etiology and the development of targeted therapies.
• The method outperforms current state-of-the-art techniques in signature discovery.