raxml-ng on Biowulf

Quick Links

RAxML-NG is a phylogenetic tree inference tool which uses maximum-likelihood (ML) optimality criterion. Its search heuristic is based on iteratively performing a series of Subtree Pruning and Regrafting (SPR) moves, which allows to quickly navigate to the best-known ML tree. RAxML-NG is a successor of RAxML (Stamatakis 2014) and leverages the highly optimized likelihood computation implemented in libpll (Flouri et al. 2014). RAxML-NG offers improvements in speed, flexibility and user-friendliness over the previous RAxML versions. It also implements some of the features previously available in ExaML (Kozlov et al. 2015), including checkpointing and efficient load balancing for partitioned alignments (Kobert et al. 2014).

Reference

RAxML-NG: a fast, scalable and user-friendly tool for maximum likelihood phylogenetic inference Alexey M Kozlov, Diego Darriba, Tomás Flouri, Benoit Morel, Alexandros Stamatakis . Bioinformatics, Volume 35, Issue 21, 1 November 2019, Pages 4453-4455, https://doi.org/10.1093/bioinformatics/btz305

Documentation

Important Notes

Module Name: raxml-ng (see the modules page for more information)
Multithreaded. MPI currently disabled on Biowulf.
Example files in /usr/local/apps/raxml-ng/examples

Interactive job

Interactive jobs should be used for debugging, graphics, or applications that cannot be run as batch jobs.

Allocate an interactive session and run the program.
Sample session (user input in bold):

[user@biowulf]$ sinteractive --cpus-per-task=4 --time=3:00:00
salloc.exe: Pending job allocation 46116226
salloc.exe: job 46116226 queued and waiting for resources
salloc.exe: job 46116226 has been allocated resources
salloc.exe: Granted job allocation 46116226
salloc.exe: Waiting for resource configuration
salloc.exe: Nodes cn3144 are ready for job

[user@cn3144 ~]$ module load raxml-ng

[user@cn3144 ~]$  cp $RAXMLNG_EXAMPLE_DATA/myoglobins597.phy .

[user@cn3144 ~]$  raxml-ng --msa myoglobin597.phy --model GTR+G --threads $SLURM_CPUS_PER_TASK
RAxML-NG v. 2.0.1 released on 24.04.2026 by The Exelixis Lab.
Developed by: Oleksiy M. Kozlov and Alexandros Stamatakis.
Contributors: Diego Darriba, Tomas Flouri, Benoit Morel, Sarah Lutteropp, Ben Bettisworth,
              Julia Haag, Anastasis Togkousidis, Julius Wiegert, Christoph Stelz.
Latest version: https://github.com/amkozlov/raxml-ng
Questions/problems/suggestions? Please visit: https://groups.google.com/forum/#!forum/raxml
...
Analysis options:
  run mode: ML tree search (adaptive)
  start tree(s): adaptive
  random seed: 1779138192
  tip-inner: OFF
  pattern compression: ON
  per-rate scalers: OFF
  site repeats: ON
  logLH epsilon: general: 10.000000, brlen-triplet: 1000.000000
  stopping rule: OFF
  fast spr radius: AUTO
  spr subtree cutoff: 1.000000
  fast CLV updates: ON
  branch lengths: proportional (ML estimate, algorithm: NR-FAST)
  FreeRate optimization method: AUTO
  SIMD kernels: AVX2
  parallelization: coarse-grained (auto), PTHREADS (4 threads), thread pinning: OFF

[00:00:00] Reading alignment from file: myoglobins597.phy
[00:00:00] Loaded alignment with 593 taxa and 6608 sites

Alignment comprises 1 partitions and 6561 patterns

Partition 0: noname
Model: GTR+FO+G4m
Alignment sites / patterns: 6608 / 6561
Gaps: 80.46 %
Invariant sites: 2.35 %


[00:00:00] Adaptive mode: Predicting difficulty of the MSA ...
[00:00:00] Parallel parsimony: 24 trees with 4 threads
[00:00:02] Predicted difficulty: 0.38

NOTE: Binary MSA file created: myoglobins597.phy.raxml.rba

[00:00:02] Generating 4 random starting tree(s) with 593 taxa
[00:00:02] Generating 9 parsimony starting tree(s) with 593 taxa
Parallelization scheme: 1 worker(s) x 4 thread(s)

Parallel reduction/worker buffer size: 1 KB  / 0 KB

[00:00:02] Data distribution: max. partitions/sites/weight per thread: 1 / 1641 / 26256
[00:00:02] Data distribution: max. searches per worker: 13

Starting ML tree search with 13 distinct starting trees

[00:00:02 -960718.286592] Heuristic: adaptive, stopping rules: off, epsilon = 10.000000
[00:00:02 -960718.286592] Initial branch length optimization
[00:00:05 -820998.403654] Model parameter optimization (eps = 10.000000)
[00:00:30 -803175.442517] FAST spr round 1 (radius: 25)
...

Optimized model parameters:

   Partition 0: noname
   Rate heterogeneity: GAMMA (4 cats, mean),  alpha: 12.814487 (ML),  weights&rates: (0.250000,0.670584) (0.250000,0.888457) (0.250000,1.067170) (0.250000,1.373789)
   Base frequencies (ML): 0.326955 0.174933 0.201353 0.296759
   Substitution rates (ML): 1.025417 1.702307 2.008450 0.624967 1.311043 1.000000

Final LogLikelihood: -495348.559145

AIC score: 993081.118291 / AICc score: 993606.346730 / BIC score: 1001181.993579
Free parameters (model + branch lengths): 1192

WARNING: Best ML tree contains 17 near-zero branches!
...
Elapsed time: 9006.222 seconds

Consumed energy: 875.736 Wh (= 4 km in an electric car, or 22 km with an e-scooter!)

[user@cn3144 ~]$ exit
salloc.exe: Relinquishing job allocation 46116226
[user@biowulf ~]$

Batch job

Most jobs should be run as batch jobs.

#!/bin/bash
set -e
module load raxml-ng

raxml-ng --msa myoglobin597.phy --model GTR+G --threads $SLURM_CPUS_PER_TASK

Submit this job using the SLURM sbatch command.

sbatch [--cpus-per-task=#] [--mem=#] [--time=DD-HH:MM:SS] raxml-ng.sh

Chaining long raxml-ng jobs

Checkpointing is built in to raxml-ng. Thus, if you have a longrunning raxml-ng job that is terminated because you did not specify a long enough walltime, or if your job is going to require more than the Biowulf 10-day max walltime, you can resubmit the job and it will restart from the last checkpoint. See the advanced tutorial for details.

You can take advantage of this feature to run a chain of jobs, each of which will pick up where the previous one terminated, using job dependencies. For example:

biowulf% sbatch --cpus-per-task=16 --mem=20g  --time=8-00:00:00 myjobscript
1111
biowulf% sbatch --depend=afterany:1111 --cpus-per-task=16 --mem=20g  --time=8-00:00:00 myjobscript
2222
biowulf% sbatch --depend=afterany:2222 --cpus-per-task=16 --mem=20g  --time=8-00:00:00 myjobscript
3333

Each of these jobs will run for 8 days. When job 1111 terminates, job 2222 will start up from the last checkpoint file, and likewise for job 3333. The 3 jobs will utilize a total walltime of 24 days.

Swarm of Jobs

A swarm of jobs is an easy way to submit a set of independent commands requiring identical resources.

Create a swarmfile (e.g. raxml-ng.swarm). For example:

raxml-ng --msa file1.phy --model GTR+G --threads $SLURM_CPUS_PER_TASK
raxml-ng --msa file2.phy --model GTR+G --threads $SLURM_CPUS_PER_TASK
raxml-ng --msa file3.phy --model GTR+G --threads $SLURM_CPUS_PER_TASK
[...]

Submit this job using the swarm command.

swarm -f raxml-ng.swarm -g 20  -t 8  --module raxml-ng

where

`-g 20`	20 Gigabytes of memory required for each process (1 line in the swarm command file)
`-t 8`	8 threads/CPUs required for each process (1 line in the swarm command file).
`--module raxml-ng`	Loads the raxml-ng module for each subjob in the swarm