Fig. 6: The relative performance of MAPLE deteriorates at higher levels of divergence.

On data simulated with varying levels of divergence (x-axis values) relative to our baseline, the relative performance of methods tailored for pandemic-scale data (MAPLE and UShER/matOptimize) decreases with higher divergence. Here we always consider a dataset of 2,000 randomly sampled sequences (Methods). a, Time required to run each method. Each phylogenetic inference method considered is represented by a different color and line style (see legend). We ran each method up to the maximum divergence level that could be analyzed due to time (1 week) and memory limitations. Each violin plot summarizes values for ten replicates, and dots represent mean values. b, Maximum RAM demand required to run each method. c, Proportional Robinson–Foulds distances between estimated trees and true trees used in simulations. Higher values correspond to more errors in phylogenetic estimation. d, Log-likelihoods (LK) (computed with IQ-TREE 2) of phylogenies inferred by different methods, relative to the highest log-likelihood score obtained by any method for the same replicate. Higher values on the y-axis represent more likely estimates.