Comparison between multi - locus and concatena - tion methods

We compared the most accurate multi-locus inference method (MLE-length) to two concatenation methods which include NeighborNet [1] and the least squares method of Schliep [2], which we refer to here as SplitsNet. We ran the concatenation methods using their default settings. The splits distance was used to evaluate the topological error, which quantifies the proportion of bipartitions that differ between the model and inferred phylogenies. As shown in Figure S1, the three methods fell into three categories based on their topological accuracy: MLE-length was the most accurate, NeighborNet was the second most accurate, and SplitsNet was the least accurate method. These results suggest that concatenation methods are less accurate than multi-locus inference methods. We also observed an increase in the topological error across all methods as the number of taxa increased from five to ten.