Fig. 5

The method’s performance is robust for different sample size and effect size. a The statistical power for each feature (the probability of including the feature into the model) plotted against sample size (other parameters as in Fig. 4 a-c). Relevant features have high probability of being included in the model (power ≥0.8) when 2000 or more transition events (corresponding to approximately 44 genealogies) are used for the analysis. Solid (dashed) lines correspond to relevant (irrelevant) features in the respective scenario. b The power for the density feature ϕ ₂ shown as a function of sample size and relative effect size. The red line indicates the section corresponding to a with a relative effect size of 1. As expected, power increases with sample size and relative effect size. c,d The power as a function of sample size and relative effect size for all c relevant and irrelevant d features of the scenario. Colorbar as in b. e-h Analogous to a-d, but for a dataset used in Fig. 4 d-f, where the transition rate λ depends on time and local cell density with a Gaussian kernel