Identification Of Active Mutational Signatures With decompTumor2Sig

Introduction:

In many cases, the somatic mutations of a tumor stem from multiple mutational processes such as cigarette smoke, UV light, or age-related spontaneous deamination of 5-methylcytosine. The identification of the processes that have contributed to an individual tumor genome is an emerging research question in cancer genomics and may be of clinical relevance, for instance, when it highlights a DNA repair deficiency which may impact the response to cytotoxic treatments.

Recently, two different mathematical models of mutational processes have been developed, describing them as "mutational signatures" in terms of mutation frequencies of altered bases within their immediate sequence context, one developed by Alexandrov et al[1] and one by Shiraishi et al[2]. These two models differ with respect to their accuracy of describing mutational events and the number of parameters required to capture the frequencies of sequence patterns altered by a mutational process.

The de novo establishment of mutational signatures requires a large number of tumor genomes[1,2] to decompose them into a (i) set of signatures that reflect the processes driving the somatic mutations across the tumors, and (ii) a set of "exposures" or contributions of the signatures to the single tumors. In a clinical setting the necessity for large set of tumors is unsatisfactory and often impractical. However, once accurate signatures have been defined, they can be used to estimate their contributions to the overall mutational load of an individual tumor sample.

Method:

We illustrate how quadratic programming can be used to determine the contribution of a given set of mutational signatures to a tumor genome. The basic idea of this optimization technique is to determine the contributions in such a way that the differences between the observed mutation frequencies in the tumor genome and the frequencies predicted by the signatures` contributions are minimal.

Discussion and conclusion:

We present the R package "decompTumor2Sig" that we explicitly developed for estimating the contribution of a given set of mutational signatures of both the Alexandrov- and the Shiraishi-type model. So far such tools existed only for Alexandrov signatures. We show the application of our tool to examples of both signature types and illustrate its effectiveness to dissect tumor genomes into a given set of mutational signatures.

References:

[1] Alexandrov et al: Signatures of mutational processes in human cancer. Nature 500, 415-421 (2013).
[2] Shiraishi et al: A simple model-based approach to inferring and visualizing cancer mutation signatures. PLoS Genet 11(12), 1005657 (2015).