De novo mutational signature discovery in tumor genomes using SparseSignatures

Abstract

Cancer is the result of mutagenic processes that can be inferred from tumor genomes by analyzing rate spectra of point mutations, or “mutational signatures”. Here we present SparseSignatures, a novel framework to extract signatures from somatic point mutation data. Our approach incorporates a user-specified background signature, employs regularization to reduce noise in non-background signatures, uses cross-validation to identify the number of signatures, and is scalable to large datasets. We show that SparseSignatures outperforms current state-of-the-art methods on simulated data using a variety of standard metrics. We then apply SparseSignatures to whole genome sequences of pancreatic and breast tumors, discovering well-differentiated signatures that are linked to known mutagenic mechanisms and are strongly associated with patient clinical features.

SparseSignatures: An R package using LASSO-regularized non-negative matrix factorization to identify mutational signatures from human tumor samples

“…While it can be omitted from the analysis, we suggest including it explicitly. The package provides two background signatures: the SBS5 signature extrapolated from the COSMIC database ( Lal et al., 2021 ), and a signature derived from the human germline mutation spectrum ( Rahbari et al., 2016 ). Import one of the available background signatures:

# To load the SBS5 signature from COSMIC data(background) # OR, for the human germline-derived signature data(background2)

…”

“…Mutational signatures are discovered by analyzing ensemble point-mutation counts from a sample of individuals. Each individual record lists the count of mutations detected in the cancer sample, classified with respect to the trinucleotide contexts: the observed nucleotide substitution (C>A, C>G, C>T, T>A, T>C, or T>G) and the neighboring 5′ and 3′ nucleotides, for a total of 96 trinucleotide mutation types ( Lal et al., 2021 ). The counts are naturally summarized in a rectangular data-frame (an R dataframe, in our case) whose row records refer to the individual cancer samples and whose columns correspond to the 96 types.…”

“…The NMF method has proved to be effective, as meaningful signatures have been identified in the literature. However, the algorithm is susceptible to discovering additional spurious signatures, that are not reflecting real prognostic factors, instead being artifacts that result from data overfitting ( Lal et al., 2021 ). For example, a subset of the mutation counts that should be classified as the same signature may be split into multiple ones.…”

“…In principle, an L1-regularization term may also be added for the vector α 0 and the matrix α , with the effect of “sparsifying” the exposure values. However, we will not discuss it in the present guide, as we have found that performing LASSO to the α values does not yield significantly improved results, while increasing the computational cost of the algorithm ( Lal et al., 2021 ).…”

“…We describe installation details and illustrate a step-by-step approach to (1) prepare the data for signature analysis, (2) determine the optimal parameters, and (3) employ them to determine the signatures and related exposure levels in the point mutation dataset. For complete details on the use and execution of this protocol, please refer to Lal et al. (2021) .…”