sigfit: flexible Bayesian inference of mutational signatures

The premise of mutational signature analysis is that genomes acquire somatic mutations through the collective action of discrete mutational processes, and its aim is to infer the mutational signatures of these processes, together with their respective levels of activity, from observed mutation data. Different models for mutational signature analysis have been developed in recent years, most prominently based on non-negative matrix factorisation (NMF). Here we present sigfit, an R package for mutational signature analysis that employs Bayesian inference to perform fitting of mutational signatures to data and extraction of mutational signatures from data, using both NMF-inspired and alternative probabilistic models. We compare the performance of sigfit to prominent existing software for mutational signature analysis, and find that it compares favourably. Furthermore, sigfit introduces novel probabilistic models to perform simultaneous fitting and extraction of mutational signatures, with the aim of powering the detection of rare or previously undescribed signatures. The package provides user-friendly data visualisation routines and is easily integrable with other bioinformatic R packages.