Data from: "DGM Is Dead; Long Live DGM-Star! A Quantization-Theoretic Approach to Gamma Rate Discretization"

dgm_star_data.tar.gz This is the main archive. Our data mirrors, where possible, Ferretti et al.'s Data from: Biased estimates of phylogenetic branch lengths resulting from the discretised Gamma model of site rate heterogeneity (see the corresponding paper). Fig. 1 (simulated trees) The figure uses Ferretti et al.’s Figs1-3-S1.tar.gz alongside newly generated data; see details below. Fig. 1a,b Ferretti’ et al.'s subfolder “Branch length effect” has simulated trees, simulated alignments, and reconstructed trees for DGM and FreeRate. We use the ones that correspond to the continuous (_cont) gamma generating model. The corresponding trees we reconstructed using DGM-Star are in fig1/fig1ab_trees_dgmstar. Fig. 1c Ferretti et al.’s subfolder “Sample size effect” has simulated trees, alignments, and reconstructed trees for DGM and FreeRate. The corresponding trees we reconstructed using DGM-Star are in fig1/fig1c/fig1c_ferretti_trees_dgmstar. In addition to Ferretti et al.’s trees/alignments for their version of Fig. 1c (see above), we simulated bigger trees to better see the saturation of the bias for all three considered rate discretization methods (DGM, FreeRate, and DGM-Star). The ground-truth trees and the script to simulate the corresponding alignments are in fig1/fig1c_big_trees_ground_truth (the exact alignments are provided separately in fig1c_big_trees_als.tar.gz, see below). The reconstructed trees for all three methods are in fig1/fig1c_big_trees_reconstructions. Fig. 1d Here we are investigating the bias at increasingly large branch lengths (beyond the range of branch lengths considered in Ferretti et al.'s Fig. 1a,b and for bigger 500-tip trees). The ground-truth trees and the script to simulate the corresponding alignments are in fig1/fig1d_ground_truth (the exact alignments are provided separately in fig1d_als.tar.gz, see below). The reconstructed trees for all three methods are in fig1/fig1d_reconstructions. Fig. 2 (HIV dataset) In fig2/IDS I provide the IDs of the sequences used for the experiments in Fig. 2 in the same format as Ferretti et al.’s PANGEA_IDS.zip. The sequences themselves are available here. The exact 1000-sequence alignment we subsampled from is available separately in fig2_aln_1000.fasta.gz. Figs. 5, 6 (TreeBASE benchmark) fig5_6_TreeBASE_benchmark_dataset_ids.txt contains the IDs of the 300 MSA datasets from TreeBASE used for Figs. 5 and 6. Fig. 7 (simulated tree with a wide range of branch lengths) The simulated 500-tip tree used for Fig. 7 is available at fig7/ground_truth_tree.nex, the corresponding AliSim-simulated alignment at fig7/simulated_MSA.fasta, and the reconstructed trees at fig7/reconstructed_{dgm,dgmstar,freerate}.raxml.bestTree. fig*_results.tar.gz archives These archives contain raxml-ng output files (including .log files) from tree reconstructions performed for the corresponding figures. fig1c_big_trees_als.tar.gz and fig1d_als.tar.gz These two archives provide the exact alignments we generated for new trees in Fig. 1C and Fig. 1D (see above). We noticed that AliSim may generate different alignments even when the same random seed is specified. This does not meaningfully affect the results; nevertheless, we provide the exact alignments here for completeness (included as standalone archives because of their size, so they do not have to be downloaded by default).