A comprehensive comparison of tools for fitting mutational signatures.

Area of Science:

• Genomics
• Cancer Biology
• Bioinformatics

Background:

• Mutational signatures link genomic patterns to cancer processes.
• Understanding these signatures aids in tumor evolution, prognosis, and therapy.
• Tools for de novo signature extraction are benchmarked, but signature fitting tools lack comprehensive evaluation.

Purpose of the Study:

• To comprehensively evaluate twelve signature fitting tools.
• To assess tool performance on synthetic and real mutational catalogs.
• To identify optimal tools for fitting known mutational signatures.

Main Methods:

• Developed synthetic mutational catalogs with empirically driven signature weights for eight cancer types.
• Evaluated twelve signature fitting tools using these synthetic catalogs.
• Assessed tool performance on real mutational catalogs.

Main Results:

• SigProfilerSingleSample performed best for samples with fewer mutations.
• SigProfilerAssignment/MuSiCal performed best for samples with more mutations.
• Constraining reference signatures led to inferior results, and absent signatures in reference catalogs posed significant challenges for all tools.

Conclusions:

• SigProfilerSingleSample and SigProfilerAssignment/MuSiCal are recommended for signature fitting based on mutation load.
• Ad hoc signature list constraints should be avoided.
• The presence of uncatalogued signatures in real cancer data remains a challenge for current tools.