Computing branch support metrics - amkozlov/raxml-ng GitHub Wiki

RBS: Rapid bootstrap (RAxML8-style)

Infer 100 rapid bootstrap replicate trees:

raxml-ng --bootstrap --msa A7.fa --model LG+G -bs-metric rbs --bs-trees 100

Run a regular ML tree search and estimate support on the best-found ML tree:

raxml-ng --all --msa A7.fa --model LG+G --bs-metric rbs

Reference: Stamatakis et al. 2008

EBG: Educated Bootstrap Guesser

Estimate branch support for a reference tree without pre-optimized model and/or branch lengths (e.g. obtained with parsimony, FastTree):

raxml-ng --ebg --msa A7.fa --model LG+G --tree reftree.nw

Estimate branch support for an existing tree with pre-optimized model parameters (e.g. from a previous raxml-ng run, similar to Python version of EBG):

raxml-ng --ebg --msa A7.fa --model A7.bestModel --tree A7.bestTree --opt-model off --opt-branches off

Run a regular ML tree search and estimate support on the best-found ML tree:

raxml-ng --all --msa A7.fa --model LG+G -bs-metric ebg

NOTE: raxml-ng integrates the lightweight version of EBG regressor with a reduced set of features. It only reports median predicted branch support. We think that this is a good tradeoff between speed, implementation complexity, and ease of result interpretation. However, if you need advanced features such as classifier, lower bound prediction, and uncertainty estimation, please use full-featured version of EBG: https://github.com/wiegertj/EBG

Reference: Wiegert et al. 2024

SH-aLRT

Estimate branch support for an existing tree, with 1000 replicates:

raxml-ng --sh --msa A7.fa --model LG+G --tree A7.bestTree --sh-reps 1000

Run a regular ML tree search and estimate support on the best-found ML tree:

raxml-ng --all --msa A7.fa --model LG+G --bs-metric sh

Reference: Guindon et al. 2010

TBE: Transfer Bootstrap Expectation

Compute transfer bootstrap from an existing set of replicate trees:

raxml-ng --support --tree bestML.tree --bs-trees bootstraps.tree --bs-metric TBE

Run a regular ML tree search and estimate support on the best-found ML tree:

raxml-ng --all --msa ali.fa --model GTR+G --bs-metric tbe

Reference: Lemoine et al., Nature 2018

PS/PBS: Parsimony (Bootstrap) Support

Generate 500 parsimony bootstrap replicate trees:

raxml-ng --bootstrap --msa A7.fa --model LG+G -bs-metric pbs --bs-trees 500

Run a regular ML tree search and estimate support on the best-found ML tree:

raxml-ng --all --msa A7.fa --model LG+G -bs-metric ps,pbs

IC/TC: Internode certainty

  1. All-in-one: run ML tree-inference, infer bootstrap replicate trees, and compute IC/TC scores as well as standard Felsenstein bootstap (FBP) supports:

raxml-ng --all --msa ali.phy --model GTR+G --bs-metric fbp,ic1 --prefix ictc_ALL

  1. Compute IC/TC and ICA/TCA scores for a given ML topology and a set of replicate trees:

raxml-ng --support --tree ML.tree --bs-trees BSREP100.tree --bs-metric ic1,ica --prefix ictc_ML

  1. Compute ICA supports on a majority-rule (MR) consensus tree:

raxml-ng --support --tree cons:MR --bs-trees BSREP100.tree --bs-metric ica --prefix ictc_MR

For more details on IC/TC please read: https://doi.org/10.1093/molbev/msu061