CMStatistics 2021: Start Registration
View Submission - CMStatistics
B0287
Title: Compositional models for mutational signature analysis Authors:  Lena Morrill - University of Cambridge (United Kingdom) [presenting]
Florian Markowetz - University of Cambridge (United Kingdom)
Abstract: Mutational processes leave their imprint on the DNA in the form of inherited mutations from ancestor cells, accumulating over time. We are interested in a particular type of mutation called a copy number change, in which big sections of the genome are gained, lost, or re-arranged. Our group previously described a method to extract copy number mutational signatures. The crucial quantities of interest are the ``signature exposures'', which indicate the fraction of copy number changes attributable to each mutational process. These quantities are compositional, in that they sum up to one sample-wise, and we can only analyse their relative abundances. We present a partial ILR mixed-effects model for non-zero exposures and a correlated Bernoulli model for zero exposures to determine if the mutational signature spectrum changes in a coordinated way between two conditions. We apply these models to address fundamental, but unanswered, questions in high grade serous ovarian carcinoma (HGSOC), one of the most deadly cancer types. We answer the question of whether early-stage HGSOC differs from late-stage HGSOC (it does), whether primary HGSOC differs from relapsed HGSOC (it does not), and how HGSOC samples with whole-genome duplication, which appears to be crucial in cancer progression, differ from diploid genomes.