RaxmlEdit

I can’t write from a particular political viewpoint. Here is a neutral encyclopedia article about RAxML.

RAxML (Randomized Axelerated Maximum Likelihood) is a widely used open-source software package for inferring phylogenies under the maximum likelihood framework. It is optimized for speed and scalability, enabling researchers to analyze large alignments of DNA, RNA, and protein sequences on multicore processors and high-performance computing clusters. Developed by Alexandros Stamatakis, RAxML has become a standard tool in many laboratories and is frequently discussed alongside other phylogenetics programs such as PhyML and MrBayes. The software is valued for its ability to deliver robust maximum likelihood estimates of evolutionary trees for large datasets, often with substantial reductions in compute time compared with earlier implementations.

RAxML’s core purpose is to estimate trees that maximize the likelihood of observing a given sequence alignment under specified evolutionary models. It supports a range of substitution models appropriate for different data types, including DNA models like GTR and associated rate heterogeneity models such as Gamma distribution for among-site rate variation. It can also handle amino acid models (for example, WAG, JTT, and LG). Users can perform comprehensive analyses that include best ML trees, as well as rapid bootstrap analyses to assess statistical support for inferred relationships. The program also accommodates partitioned analyses, where different data blocks (e.g., genes or codon positions) are allowed to have separate model parameters.

Overview

  • Purpose and scope: RAxML is designed for inference of phylogenetic trees under maximum likelihood, accommodating large-scale datasets typical of modern phylogenomics projects. It is commonly used in evolutionary biology, systematics, and related fields to explore relationships among species, genes, or other biological units. See Phylogenomics for broader methodological contexts.
  • Data types and models: The software supports nucleotide and amino acid sequences and a broad set of substitution models. Users select models and rate variation schemes to fit their data, with model choice affecting tree estimation and interpretation. See Substitution model (phylogenetics) and Maximum likelihood for foundational concepts.
  • Outputs: RAxML reports a best-scoring ML tree, bootstrap trees, and associated support metrics. Results can be interpreted in light of model assumptions, data quality, and the biological question under study.

Algorithms, models, and heuristics

  • Maximum likelihood framework: The central aim is to identify a tree topology and model parameters that maximize the probability of observing the given sequence data. See Maximum likelihood for the statistical underpinnings.
  • Tree search strategies: To navigate the vast space of possible tree topologies, RAxML employs heuristic search methods that balance accuracy and computational efficiency. Common techniques include subtree pruning and regrafting (SPR) and near-neighbor interchange (NNI) moves, often used in combination with rapid hill-climbing. See Subtree pruning and regrafting and Heuristic search.
  • Bootstrap and support assessment: Rapid bootstrap analysis is a key feature, enabling users to gauge support for clades without excessive computational burden. The software implements strategies to determine when bootstrap iterations are sufficient, which is important for interpreting confidence in inferred relationships. See Bootstrapping (statistics).
  • Data partitioning and mixed models: In partitioned analyses, different parts of the alignment can be modeled with distinct substitution processes, allowing for more realistic handling of heterogeneity across genes or regions. See Partitioned analysis (phylogenetics).

Implementation, performance, and accessibility

  • Parallelization and hardware use: RAxML is designed to exploit parallel architectures, including multi-core CPUs (via threaded execution) and MPI-based clusters, to accelerate tree inference and bootstrap analyses. This makes it suitable for very large datasets common in modern phylogenetics. See Parallel computing.
  • Software lineage and variants: The core RAxML program has undergone multiple major updates, with RAxML-NG as a later, next-generation iteration that expands capabilities, improves scalability, and broadens model support. See RAxML-NG for details.
  • Input formats and workflows: The program accepts common sequence alignment formats and interfaces with standard phylogenetics workflows, often serving as a backbone for large-scale analyses that feed into downstream interpretation and visualization tools. See Multiple sequence alignment and Phylogenetic tree.

Controversies and methodological debates

  • Concatenation versus species tree approaches: In multilocus data sets, researchers debate whether concatenating genes and inferring a single ML tree (as RAxML typically does in its standard workflow) accurately reflects species history, given processes like incomplete lineage sorting. Proponents of concatenation emphasize practical benefits and often robust results in many datasets, while critics highlight potential biases and advocate for species-tree methods that explicitly model gene-tree discordance (e.g., coalescent theory and species tree methods). See discussions around Coalescent theory and Phylogenomics debates.
  • Model misspecification and data quality: Like all likelihood-based methods, RAxML’s inferences are sensitive to model choice and data quality. Critics note that poor model fit or systematic biases in alignment can bias tree estimates, underscoring the importance of model testing, data curation, and complementary analyses. See Model selection.
  • Heuristic limitations: Because RAxML relies on heuristics to search a huge space of trees, there is no guarantee of finding the absolute ML tree for very large or complex datasets. This has led to ongoing methodological discussions about search strategies, initialization, and convergence assessment. See Heuristic search.

See also