a) Stacked bar chart showing the number of clonal mutations (present in all tumour cells, light blue) and subclonal mutations (dark blue) in each patient. Also shown is the percentage of subclonal variants (orange triangles). For patients with multiple samples, the fraction of tumour cells used was derived from the earliest sample where the mutation was found. ( b– d) Statistical modelling by a Bayesian Dirichlet process of the distribution of clonal and subclonal mutations for three patients for which only one sample was available. For each plot, the faction of tumour cells carrying the variant is represented on the x axis (1=100% of tumour cells), and the probability density (on an arbitrary scale) on the y axis. Grey bars represent the histogram of mutations, with the fitted distribution as a dark purple line. The 95% posterior confidence intervals for the fitted distribution are represented by a pale blue area. ( b) Patient showing a vast majority of clonal variants; ( c) patient with a dominant set of clonal mutations and a minor subclone; ( d) patient showing a dominant set of subclonal mutations, at two different proportions. ( e) Adjusted fraction of tumour cells carrying KRAS, NRAS and BRAF substitutions found in the study. Error bars represent 95% confidence intervals, accounting for chromosomal copy number of the locus, percentage contaminating normal cells and depth of coverage. (Note that patient PD5883, carrying a BRAF indel, was not included in this panel due to the inaccurate estimation of the allelic fraction of indels). All mutations whose confidence interval includes 1 (red line) are considered to be present in all tumour cells with 95% confidence. Driver mutations can be found at the clonal or subclonal level, and sometimes coexist in the same patient (Patient IDs in red). For patients with multiple samples, the fraction of tumour cells used was derived from the earliest sample where the mutation was found. In three patients (PD5876, PD5885, PD5892), a fraction of ~1.5 was assigned, likely due to focal subclonal gains of the variant locus that went undetected by our algorithms, so that the estimated fraction of tumour cells was not adjusted.