Mutational Signature Analysis

Mutational signature contributions were inferred using deconstructSigs (Rosenthal et al., 2016) and the choice of signatures was further informed using results from SigProfiler (Alexandrov et al., 2013b). Only samples with at least 50 mutations were employed in the analysis, for a total of 6,410 samples.

SigProfiler was used to infer mutational signatures from TCGA whole-exome sequencing data. For each TCGA study of interest, input mutational matrices were generated using the SigProfilerMatrixGeneratorR function using all samples containing at least 50 mutations and aligning the MAF files to the hg38 genome build. SigProfilerExtractorR was used to extract de novo mutational signatures for each cancer type. For each study, the solution with the greatest number of mutational signatures was chosen, for which also the sum of the solution stability (calculated average silhouette coefficient) and minimum stability exceeded 1, and the minimum stability value did not fall below 0.4. For the selected solutions, the identity of the mutational signatures was determined by calculating cosine similarities with COSMIC v3.1 mutational signatures.

Mutational signatures from whole-exome sequencing data were also inferred using the deconstructSigs R package. MAF files were aligned against the hg38 genome build and the COSMIC v3.1 mutational signatures were employed in the analysis. For each cancer type, the contribution of all signatures within the deconstructSigs solution was set to 0, apart from SBS1, SBS5, signatures identified in the corresponding SigProfiler solution, as well as signatures which contribute on average at least 5% of mutations across all samples within the deconstructSigs solution. For cancer types where SigProfiler did not result in a stable solution, the deconstructSigs solution was used with the contribution of all signatures set to 0, apart from SBS1, SBS5 as well as signatures which contribute at least 5% of mutations across samples within the deconstructSigs solution.