Hidden Markov Models Lead to Higher Resolution Maps of Mutation Signature Activity in Cancer

Abstract

Abstract. Knowing the activity of the mutational processes shaping a cancer genome may provide insight into tumorigenesis and personalized therapy. It is thus important to uncover the characteristic signatures of active mutational processes in patients from their patterns of single base substitutions. However, mutational processes do not act uniformly on the genome and are biased by factors such as the genome's chromatin structure or replication origins. These factors may lead to statistical dependencies among neighboring mutations, calling for modeling approaches that can account for such dependencies to better estimate mutational process activities.Here we develop the first sequence-dependent models for mutation signatures. We apply these models to characterize genomic and other factors that influence the activity of previously validated mutation signatures in breast cancer. We find that our tool, SigMa, can accurately assign genomic mutations to mutation signatures, yielding assignments that are of higher likelihood than those obtained with models that assume independence between signatures and align better with current biological knowledge. Our analysis resolves a controversy related to the dependency of APOBEC signatures on replication time and links Signatures 18 and 30 to oxidative damage.Modeling the sequential dependencies of mutation signatures leads to improved estimates of mutation signature activity both at the tumor-level and within specific genomic regions, yielding higher resolution maps of mutation signature activity in cancer.

cancerAlign: Stratifying tumors by unsupervised alignment across cancer types

“…To better understand cancer heterogeneity, large-scale cancer genomics projects, such as the Cancer Genome Atlas (TCGA), the International Cancer Genome project, and the Memorial Sloan Kettering-Integrated Mutation Profiling project have systematically profiled thousands of tumors, holding the promise to realize personalized treatment [3][4][5][6]. Among the collected genome-scale omics data, somatic mutation profiles have been used to discover causal drivers of tumors [7][8][9][10][11][12][13] and further reveal informative cancer subtypes [14,15]. In pursuit of this vision, computational approaches have been developed to stratify tumors according to high-dimensional, noisy and sparse somatic mutation profiles [1,[16][17][18][19][20][21].…”