Software Choice Matters in Mutational Signature Refitting

Background: Mutational signature refitting is widely used to quantify the contributions of known mutagenic processes in cancer genomes. However, different software implementations can yield discordant results even when applied to the same data. Methods: This study compares five commonly used refitting tools—Sigminer, MutationalPatterns, sigfit, musicatk, and SigProfilerAssignment—across two independent datasets: TRACERx whole-exome sequencing data and a Genome in a Bottle (GIAB) reference sample. Inter-tool agreement was evaluated using six complementary metrics: Frobenius distance, Manhattan distance, cosine similarity, Pearson correlation, covariance-based distance, and spectral distance. Results: Across all metrics, musicatk and Sigminer demonstrated the strongest concordance, while sigfit was the most divergent. This ranking was consistent in the single-sample GIAB analysis, suggesting that the observed discrepancies arise primarily from implementation differences between tools rather than from dataset-specific factors. Conclusions: Our findings underscore the importance of cross-tool validation and careful consideration of tool-dependent variability when mutational signatures are used for biological interpretation or biomarker development.