Fig. 3

Marked improvement in mutational signature analysis explicability in the TCGA-HNSC samples using FIREVAT. a, b Each panel is comprised of the following plots from top to bottom: distribution of signature weights for the TCGA-HNSC samples (n = 130), bar plot of the number of mutations in each sample, histogram of cosine similarity scores from signature analysis, and correlation between the sum of tobacco signature weights and the number of pack-years among current smokers. In the two plots of signature weights, the green bars indicate the contribution weights of smoking-related signatures in each sample while the dark red bars represent that of artifactual signatures. a Mutational signature analysis without variant refinement. Of the 130 TCGA-HNSC samples, substantially high levels of artifactual signature weights were identified (median weight sum = 45.3%, min = 3.2%, max = 100%). The Pearson correlation between the sum of tobacco signatures and the number of pack-years was negligent using an unrefined variant list (r = 0.094). In particular, one sample had somatic hypermutations (15.6 mutations/Mb; denoted with an asterisk). b Mutational signature analysis with variant refinement by FIREVAT. Compared to the unrefined callset, the correlation between the sum of tobacco signature weights and the number of pack-years was higher (r = 0.23) and the weights of artifactual signatures were decreased (median weight sum = 0%, min = 0.0%, max = 30.6%). c, d Unveiling biologically relevant mutational signatures by removing mutations of artifactual signatures. c Mutation frequency spectrum of unrefined, refined, and artifactual mutations from the case TCGA-CR-7399 (HNSC) and SBS45 (8-oxoG signature). In the spectrum plot of refined and artifactual mutations, the asterisks represent frequency peaks found in different signatures (green = SBS4, orange = SBS2 and SBS13, red = SBS43, SBS45, SBS49, and SBS53). d Mutational signature weights of unrefined, refined, and artifactual mutations from TCGA-CR-7399. The tobacco smoking and APOBEC-related signatures were identified only from the signature analysis results of FIREVAT-refined mutations