Computational approaches for discovery of mutational signatures in cancer

Abstract

The accumulation of somatic mutations in a genome is the result of the activity of one or more mutagenic processes, each of which leaves its own imprint. The study of these DNA fingerprints, termed mutational signatures, holds important potential for furthering our understanding of the causes and evolution of cancer, and can provide insights of relevance for cancer prevention and treatment. In this review, we focus our attention on the mathematical models and computational techniques that have driven recent advances in the field.

Figure 1

Mathematical modelling and deconvolution of mutational signatures. (A) Diagram illustrating the modelling of mutational signatures as probabilistic relationships between mutation types and mutational processes operative in genomes, for a general case with K mutation types, N mutational processes and G genomes. The notation of signatures, exposures and mutational catalogues follows that used in the main text. The varying widths of the links between mutation types and signatures (mutation probabilities), and between signatures and catalogues (signature exposures) represent the observation that varying values of ${\mathrm{s}}_{\text{kn}}$ and ${\mathrm{e}}_{\text{ng}}$ reflect the specific mutational profile of each signature and the exposure composition of each genome. Non-negativity constraints for mutation probabilities and signature exposures are specified directly below them. (B) Example of de novo signature extraction, for a case with K = 6 mutation (single-base substitution) types, N = 3 mutational signatures and G = 4 mutational catalogues. Starting from the set of catalogues (depicted here as mutational profiles, each bar corresponding to a distinct mutation type), de novo extraction methods determine the set of mutational signatures (represented as consensus mutational profiles) and exposures (depicted here as proportions of the mutations in each catalogue, for simplicity) that reconstruct the original mutational catalogues with minimal error.