Fig. 3

Analysis of the model in Fig. 2 using \(n=200\) non-parametric bootstrap resamples. a Bootstrapped model frequency. The target model has \(K=3\) Beta clusters, plus one tail. Across all the 200 resamples the target model is fit in 65% of the cases, and in the remaining 35% only two Beta components are used; in all cases the tail is always fit to data. b Bootstrapped distribution of the tail shape, with annotated point estimate (dashed line). c Bootstrapped distribution of the mean and variance of the Beta components, with annotated point estimate (dashed line). Note that cluster C2 can show bimodal distributions; this is due to the 35% bootstrapped cases in which C2 is the only subclonal cluster (i.e., C3 is dropped). Point estimates are annotated as dashed lines. d Co-clustering probability of each pair of points in the input data, visualized as a matrix ordered by the clustering assignments in Fig. 2. This shows that clonal mutations are very stable (C1), and that mutations in cluster C3 in a small proportion of cases are assigned together with mutations in cluster C2. This is confirmed by the plot in panel (a) where the model with \(K=2\) Beta clusters plus one tail are identified in 35% of the non-parametric bootstrap resamples