Simulation scripts and results for the guide "IQ-TREE vs RAxML-NG vs FastTree" (https://proteiniq.io/guides/iqtree-vs-raxml-vs-fasttree), run on 2026-11-12 and 2026-11-13. Programs (one core each, WSL Ubuntu on a laptop CPU, several single-core jobs sharing the machine) IQ-TREE 3.1.1 https://github.com/iqtree/iqtree3/releases/download/v3.1.1/iqtree-3.1.1-Linux.tar.gz RAxML-NG 2.0.2 https://codeberg.org/amkozlov/raxml-ng/releases/download/2.0.2/raxml-ng_v2.0.2_linux_x86_64.zip sha256 af8bd88473dfe0078323fc59f7b90bd7636f6abeef704d049baf2f3e249f904d FastTree 2.2.0 FastTree.c at commit 29c5e62fbcd93230ee325f9c6a17b81f00e3c72a, sha256 975202a6b74c9996af871404ff043bb2152edcbda539035662514bc12d1f3431, gcc -O3 -finline-functions -funroll-loops -o FastTree FastTree.c -lm Simulation (IQ-TREE AliSim) DNA: GTR{1.0,4.0,0.8,1.1,3.5,1.0}+F{0.3,0.2,0.2,0.3}+G4{0.7}, random Yule-Harding tree (RANDOM{yh/N}), -af fasta Protein: LG+G4{0.8} bench.sh NAME KIND NTAXA NSITES SEED BRANCH_SCALE [support] simulates one alignment and runs every program. Groups: acc = 100 DNA sequences x 1000 columns, seeds 21 to 25; prot = 100 sequences x 400 amino acids, seeds 31 to 33; scale = 50, 100, 200, 400 DNA sequences x 1000 columns, seed 11. IQ-TREE ModelFinder runs on the protein alignments were stopped after several hours and are not reported. The RAxML-NG default run at 400 sequences did not finish. sup2.sh SEED builds the support calibration set: 100 DNA sequences x 300 columns, --branch-scale 0.5, seeds 41 to 48. Commands FastTree FastTree -nt -gtr -gamma aln.fa (protein: -lg -gamma) IQ-TREE fixed iqtree3 -s aln.fa -m GTR+G4 -nt 1 -seed 1 (protein: LG+G4) IQ-TREE default iqtree3 -s aln.fa -nt 1 -seed 1 (ModelFinder Plus) RAxML-NG default raxml-ng --search --msa aln.fa --model GTR+G4 --threads 1 --seed 1 RAxML-NG one tree raxml-ng --search1 --msa aln.fa --model GTR+G4 --threads 1 --seed 1 Support iqtree3 -s aln.fa -m GTR+G4 -B 1000 -nt 1 -seed 1 raxml-ng --all --msa aln.fa --model GTR+G4 --threads 1 --seed 1 --bs-trees 100 FastTree default output (SH-like local support) Scoring every tree was re-scored with: iqtree3 -s aln.fa -te TREE -m GTR+G4 -nt 1 -seed 1 (LG+G4 for protein), so likelihoods are comparable across programs. Files results.tsv one row per program and dataset; seconds are wall-clock; normalized_RF is the symmetric difference of unrooted splits divided by 2(n-3); true-tree rows give the true tree's score support_branches.tsv every internal branch of every program's tree in the support set: its support value and whether the branch is in the true tree rf.py, aggregate.py, agg_support.py RF distance, summary tables, support tables example_raxml-ng_run_log.txt log of one RAxML-NG 2.0.2 bootstrap run