Fig. 3: LiFT allows classification of fit variants <2% VAF.

a, Schematic of LiFT algorithm. LiFT compares a model of clonal dynamics (Fig. 1a) with an artifact model and performs Bayesian model selection. The subsequent steps to infer clonal structure and fitness distributions are as in Fig. 1a. b, Gradient in VAF versus VAF for variants detected in the LBCs with at least two timepoints and at least one VAF > 1% per trajectory, with filtered (orange), fit (blue) and synonymous (light green dots) mutations, classified by LiFT on a logarithmic scale. c, Longitudinal trajectories of fit (blue) and filtered (orange) mutations linked to age in years. d, Number of trajectories classified as fit by LiFT, broken down into increasing or decreasing VAF from first to last timepoint. e, Left, deterministic fit of all mutations selected by LiFT in an individual of the LBC cohorts using the inferred optimal clonal structure (Supplementary Information Methods, Appendix B). 90% CIs associated with binomial sampling noise are shown for each data point. VAF is displayed on a logarithmic scale. Right, posterior distribution of fitness associated to each clonal structure. f, Fitness effects of variants broken down by filtering method. The sample size, n, and statistical analyses comparing the distribution of fitness, computed using the non-parametric Kruskal–Wallis test, are highlighted (*H = 14, P = 1 × 10⁻⁴; **H = 18, P = 1 × 10⁻⁵). g, Fitness of variants selected as fit by LiFT broken down by their maximum VAF, >2% and <2%, and damage prediction. The top row displays a bar plot of variant counts for each category. The bottom row displays box plots showing the median and interquartile range of the distribution of MAP fitnesses by damaging prediction displayed on a logarithmic scale to emphasize relative differences in fitness between variants. Consequently, of a total of 89 variants with a damage prediction, 17 variants with fitness below 2% are not shown but are reported in Supplementary Tables 4–6. A marginal plot shows the Gaussian kernel density estimation of the MAP fitness values. fs, frameshifts; ter, terminations.