Skip to content

Review: AliSim: A Fast and Versatile Phylogenetic Sequence Simulator for the Genomic Era

Citation

  • Ly-Trong, N., Naser-Khdour, S., Lanfear, R., & Minh, B. Q. (2022). AliSim: a fast and versatile phylogenetic sequence simulator for the genomic era. Molecular Biology and Evolution, 39(5), msac092.
  • DOI

Abstract

AliSim is a flexible phylogenetic sequence simulator integrated into IQ-TREE 2. It supports a wide range of substitution models, partitioned analyses, insertion and deletion events, rate heterogeneity, and codon models. Benchmarks show that AliSim is substantially faster than comparable simulators while supporting a broader range of evolutionary models.

Generating synthetic sequence alignments with known evolutionary parameters is essential for testing and validating phylogenetic methods. AliSim, integrated into IQ-TREE 2, simulates alignments along a given tree under any substitution model that IQ-TREE 2 can analyze: DNA models (JC, HKY, GTR, and many others), protein models (JTT, WAG, LG, and empirical mixtures), codon models, and partitioned analyses where different alignment regions evolve under different models.

The simulation process follows the standard Markov model: starting from a character drawn from the stationary distribution at the root, each branch generates a new character by drawing from the transition probability \( P(t)_{ij} \) for that branch length. Site rate heterogeneity under the Gamma model follows the discrete approximation of Yang (1994): each site is assigned a rate drawn uniformly from the \( K \) discrete categories, and that rate multiplies all branch lengths for that site.

The test-data generator distributed with Hifuku (scripts/generate_test_dataset.py) and the demonstration notebooks use AliSim to build small synthetic datasets with controlled properties: known tree topologies (a base tree and NNI-perturbed variants), a chosen substitution model, and a set alignment length. These datasets give the map-sanity and GPU-versus-CPU parity tests a landscape whose structure is known in advance. A user may run or adapt the generator for their own questions.