Skip to content

Review: ModelFinder: Fast Model Selection for Accurate Phylogenetic Estimates

Citation

  • Kalyaanamoorthy, S., Minh, B. Q., Wong, T. K. F., von Haeseler, A., & Jermiin, L. S. (2017). ModelFinder: fast model selection for accurate phylogenetic estimates. Nature Methods, 14(6), 587–589.
  • DOI

Abstract

ModelFinder is a fast model selection algorithm for phylogenetic analyses that evaluates hundreds of candidate substitution models using a likelihood ratio test or information criterion (AIC, AICc, BIC). It is substantially faster than previous methods while producing equivalent or better model selections, and is integrated into IQ-TREE 2.

Choosing the wrong substitution model systematically biases branch length estimates and can lead to incorrect topologies. Model selection identifies which of many candidate models (JC, K2P, HKY, GTR, and combinations with Gamma rate variation, invariant sites, and so on) best fits the data under a statistical criterion such as the Bayesian Information Criterion (BIC).

ModelFinder speeds up model selection by evaluating candidate models on a neighbor-joining tree rather than re-running full maximum-likelihood optimization for each candidate. The BIC score for a model \( M \) is:

\[\mathrm{BIC}(M) = -2 \log L(M) + k \log n\]

where \( k \) is the number of free parameters and \( n \) is the number of alignment sites. The model with the lowest BIC is selected.

Hifuku takes a caller-supplied substitution model and builds its \( Q \) matrix and eigendecomposition for the likelihood. The implemented models are JC69 and GTR for DNA and LG, WAG, and JTT for proteins; the caller passes the corresponding constructor (for example jc69() or lg()).

Future work. Automatic model selection, choosing the best-fit model for each alignment by a criterion such as the BIC score above, is a planned addition. ModelFinder is the standard tool for that step. Until it is integrated, the user picks the model upstream of Hifuku.