Extended Data Fig. 4: Clonal Dynamics and Inference - Supporting Material.

. a. Approximation of a neutral birth-death model using the negative binomial distribution. The exact model assumes symmetric divisions occur every 40 weeks, or 1.3 divisions per year, and has no bias towards self-renewal (see SI methods Section 1). b. Deterministic trajectory (see SI methods Appendix B) with maximum a posteriori (MAP) fitness and fitted time of mutation (left) and joint posterior distribution of fitness and number of wild-type HSPCs population (right) inferred from an individual with a single mutation selected by LiFT. 90% confidence intervals associated with binomial sampling noise are shown for each data point. Note that VAF is displayed on a logarithmic scale to highlight relative differences and the initial exponential growth of clones. Also note that a small random horizontal jitter has been added to data points to avoid overlapping of confidence intervals. c. Deterministic trajectory (see SI methods Appendix B) with maximum posterior fitness and fitted time of mutation (left) and joint posterior distribution of fitnesses and number of wild-type HSPCs inferred from an individual with three mutations, selected by LiFT, occurring in two clones. 90% confidence intervals associated with binomial sampling noise are shown for each data point. Note that VAF is displayed on a logarithmic scale to highlight relative differences and the initial exponential growth of clones. Also note that a small random horizontal jitter has been added to data points to avoid overlapping of confidence intervals. d. Evidence supporting the clonal structure selected by our Bayesian model comparison relative to the model assuming no mutations co-occur on the same clone. The evidence is only shown for non-trivial cases where more than one mutation was selected by LiFT in an individual.