RAxML-NG 2: Automatic model selection, novel tree search heuristics, and fast branch support metrics
- 1Heidelberg Institute for Theoretical Studies
Abstract
RAxML-NG is a widely used tool for maximum likelihood based phylogenetic inference. In the seven years since the last RAxML-NG publication, we have continuously improved and extended the code. Here, we describe the next major release, RAxML-NG 2.0. It introduces a plethora of new features: integrated model testing, multiple fast branch support metrics, automatic parallelization tuning, phylogenetic difficulty prediction, genotype evolution models, to name but the most important ones. Furthermore, we introduce two novel search heuristics at production code level: the adaptive difficulty-aware heuristic (default) and the fast mode with early-stopping that prevents over-optimization. We perform extensive benchmarking of RAxML-NG 2.0 with respect to its accuracy and speed, and compare it to other popular maximum likelihood based phylogenetic inference tools (IQTree, VeryFastTree) as well as to preceding RAxML-NG versions. In particular, the new fast search heuristic in conjunction with machine learning based branch support prediction induces a 75x inference time reduction compared to RAxML-NG 1.2, with minor to no accuracy loss. The code is available under GNU GPL at https://codeberg.org/amkozlov/raxml-ng.
Related
- Fixing the Measurement Model Before the Selection Heuristics: Optimiser Repairs for Living-Cell Self-Driving Laboratories
- Selection-aware metrics provide complementary decision-level information in plant genomic prediction
- Hit Selection Using SSMD-Based Machine Learning Performance Metrics in High-Throughput Screening Assays
- Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature
- The (implicit) force is with you: an application of the implicit function theorem for automatic protein force fields reparametrization
BibTeX
@misc{kozlov2026raxmlng,
title = {{RAxML-NG 2: Automatic model selection, novel tree search heuristics, and fast branch support metrics}},
author = {O. M. Kozlov and A. Togkousidis and C. Stelz and D. Hoehler and J. Wiegert and A. Stamatakis},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.09.750097},
url = {https://www.biorxiv.org/content/10.64898/2026.09.09.750097v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - RAxML-NG 2: Automatic model selection, novel tree search heuristics, and fast branch support metrics
AU - O. M. Kozlov
AU - A. Togkousidis
AU - C. Stelz
AU - D. Hoehler
AU - J. Wiegert
AU - A. Stamatakis
PY - 2026
DA - 2026/09/09
PB - bioRxiv
DO - 10.64898/2026.09.09.750097
UR - https://www.biorxiv.org/content/10.64898/2026.09.09.750097v1
AB - RAxML-NG is a widely used tool for maximum likelihood based phylogenetic inference. In the seven years since the last RAxML-NG publication, we have continuously improved and extended the code. Here, we describe the next major release, RAxML-NG 2.0. It introduces a plethora of new features: integrated model testing, multiple fast branch support metrics, automatic parallelization tuning, phylogenetic difficulty prediction, genotype evolution models, to name but the most important ones. Furthermore, we introduce two novel search heuristics at production code level: the adaptive difficulty-aware heuristic (default) and the fast mode with early-stopping that prevents over-optimization. We perform extensive benchmarking of RAxML-NG 2.0 with respect to its accuracy and speed, and compare it to other popular maximum likelihood based phylogenetic inference tools (IQTree, VeryFastTree) as well as to preceding RAxML-NG versions. In particular, the new fast search heuristic in conjunction with machine learning based branch support prediction induces a 75x inference time reduction compared to RAxML-NG 1.2, with minor to no accuracy loss. The code is available under GNU GPL at https://codeberg.org/amkozlov/raxml-ng.
ER -