Phylogenetic reconstruction accuracy in simulation 6 (“NGS-large”). Mutations were introduced according to signature S1, and the sequencing depth was 5x. Read counts were simulated with a 5% amplification error, 1% sequencing error, and 10% ADO. TNT, SiFit, and CellPhy-ML16 use the inferred genotypes, CellPhy-GL16, uses genotype likelihoods, and SCIPhI and ScisTree use read counts. Phylogenetic accuracy is 1—nRF (see the “Methods” section). For more than 1000 SNVs or more than 100 cells, only ten replicates were run for SiFit for reasonable running times. The (light) orange stars indicate that we could not obtain SCIPhI and ScisTree results for the largest datasets after running each replicate for >100 h. CellPhy was run using the ML genotypes error model (CellPhy-ML) and genotype likelihoods (CellPhy-GL). See Fig. 1 for an explanation of the boxplots.