DNA mutational clustering

We tested for true presence or absence of the single nucleotide variants using an approach previously described.⁵⁴ Briefly, counts at each loci were calculated across all samples using AlleleCounter (https://github.com/cancerit/alleleCount). For each patient, the non-tumour samples in this study not belonging to that patient were used as a reference to obtain the locus-specific error rate. To minimize the false positive rate, the presence of the variant in the sample was accepted if the multiple-testing corrected p-value was less than 0.001.

Mutations were clustered using a Bayesian Dirichlet based algorithm as described previously.¹⁹ Briefly, the expected number of reads for a given mutation present in one allelic copy of 100% of tumour cells may be estimated based upon the ASCAT derived tumour cell fraction, the copy number at that locus, and the total read-depth. The fraction of cells carrying a given mutation is modelled by a Dirichlet process with an adjustment for the decreased sensitivity in identifying mutations in lower tumour fractions. Mutations were thus assigned to clusters according to the calculated fraction of clonality. The hierarchical ordering of these clusters was determined by applying the pigeonhole principle.