Fig. 2: Benchmarking results for lineage reconstruction methods on TedSim simulated datasets with dropouts.

For all boxplots shown in the figure, we adopt the following default settings in the ggplot2 library: center line, median; Upper and lower box limits, the 25th and 75th percentiles; whiskers, 1.5× interquartile range for both upper and lower ends. Source data are provided as a Source Data file. a–c Comparisons of LinRace (LinRace-IST and LinRace-TST) and other methods on TedSim simulated datasets using RF distance, CID. and Nye similarity on datasets with 1024 cells. RF distance, Nye similarity, and CID all have the range of [0, 1]. For both RF distance and CID, lower is better, and for Nye similarity, higher values indicate better performance. The detailed descriptions for simulation and method settings can be found in “Methods” and Supplementary Note 2. d Comparisons of LinRace (LinRace-IST and LinRace-TST) and other methods on TedSim simulated datasets using RF distance on datasets with 4096 cells. Startle-NNI is excluded from this analysis due to computational time on this larger dataset. e RF distance comparisons of LinRace’s improvements upon different methods for building tree backbone including NJ, Cassiopeia-hybrid, and DCLEAR-kmer. In these plots, we used the same datasets (with dropouts) as in (a–c). Comparisons using Nye Similarities and CID are included in Supplementary Fig. 5.