RAxML

RAxML performs maximum-likelihood phylogenetic inference and post-analysis of large phylogenies for biological and medical research, including datasets produced by next-generation sequencing.

Key Features:

• Maximum likelihood inference: Implements maximum-likelihood phylogenetic inference with support for sequential and parallel processing.
• Substitution models and data types: Provides a comprehensive extension of substitution models and supported data types for diverse phylogenetic datasets.
• Vector intrinsics: Integrates SSE3, AVX, and AVX2 vector intrinsics to optimize computational performance.
• RAxML-VI-HPC optimizations: RAxML-VI-HPC introduces low-level technical optimizations and a modified search algorithm.
• GTR+CAT approximation: Uses the GTR+CAT approximation instead of GTR+Gamma to accelerate analyses, achieving reported speedups of 2.7 to 52 times.
• Parallelization and MPI: Employs MPI-based parallelization enabling concurrent bootstrapping and inferences from distinct starting trees.
• Memory and performance comparisons: Reported to require at least 5.6 times less main memory than GARLI, PHYML, IQPNNI, and MrBayes while producing better-quality trees within similar timeframes and running 2–3 times faster than GARLI on datasets ≥4000 taxa.
• Scalability on empirical datasets: Demonstrated on large alignments, including 25,057 taxa (1463 bp) and 2,182 taxa (51,089 bp).
• Post-analysis operations: Provides operations for conducting post-analyses on sets of phylogenetic trees.

Scientific Applications:

• Large-scale phylogenetic reconstruction: Reconstruction of phylogenies from large datasets, including those produced by next-generation sequencing.
• Bootstrap analyses: Concurrent bootstrapping for assessment of phylogenetic support.
• Benchmarking and methodological comparisons: Comparative performance and memory benchmarking against GARLI, PHYML, IQPNNI, and MrBayes.
• Post-phylogeny analyses: Post-analysis of sets of phylogenetic trees for downstream comparative and summary analyses.

Methodology:

Computational approach is based on maximum likelihood principles, the GTR+CAT approximation (versus GTR+Gamma), sequential and MPI-parallel processing, SSE3/AVX/AVX2 vector intrinsics, low-level technical optimizations, and a modified search algorithm as implemented in RAxML-VI-HPC.