Identifying the best approximating model in Bayesian phylogenetics: Bayes factors, cross-validation or wAIC?
Cite this dataset
Lartillot, Nicolas (2023). Identifying the best approximating model in Bayesian phylogenetics: Bayes factors, cross-validation or wAIC? [Dataset]. Dryad. https://doi.org/10.5061/dryad.j9kd51cfq
Abstract
There is still no consensus as to how to select models in Bayesian phylogenetics, and more generally in applied Bayesian statistics. Bayes factors are often presented as the method of choice, yet other approaches have been proposed, such as cross-validation or information criteria. Each of these paradigms raises specific computational challenges, but they also differ in their statistical meaning, being motivated by different objectives: either testing hypotheses or finding the best-approximating model. These alternative goals entail different compromises, and as a result, Bayes factors, cross-validation and information criteria may be valid for addressing different questions. Here, the question of Bayesian model selection is revisited, with a focus on the problem of finding the best-approximating model. Several model selection approaches were re-implemented, numerically assessed and compared: Bayes factors, cross-validation (CV), in its different forms (k-fold or leave-one-out), and the widely applicable information criterion (wAIC), which is asymptotically equivalent to leave-one-out cross validation (LOO-CV). Using a combination of analytical results and empirical and simulation analyses, it is shown that Bayes factors are unduly conservative. In contrast, cross-validation represents a more adequate formalism for selecting the model returning the best approximation of the data-generating process and the most accurate estimates of the parameters of interest. Among alternative CV schemes, LOO-CV and its asymptotic equivalent represented by the wAIC, stand out as the best choices, conceptually and computationally, given that both can be simultaneously computed based on standard MCMC runs under the posterior distribution.
Usage notes
- a multiple sequence alignment of elongation factor 2 in 30 eukaryotic species (627 aligned positions), taken from Lartillot and Philippe, 2006. Syst Biol 55:195.
- a concatenation of genes (35371 aligned positions) across 35 metazoans + 2 choanoflagellates and 12 fungi, originally published in Philippe et al, 2005. Mol Biol Evol 22:1246
- simulation replicates produced for table 1 (normal model).
- cross-validation replicates of the elongation factor data set used for table 2.
- jackknife replicates of the metazoa dataset used for figure 2.
- simulation replicates obtained by posterior predictive simulations used for figure 3.
- jackknife replicates of the metazoa dataset used for figure 4.
- scripts for reproducing the analyses of table 2, figures 2-4.
- source code and scripts for running the multivariate normal model of table 1 and figure 1. also available at https://github.com/bayesiancook/normcv.git.
- source code of phylobayes mpi, version 1.9, used for the phylogenetic analyses presented in the article. also available at https://github.com/bayesiancook/pbmpi.git.
The alignments are standard amino-acid multiple-sequence alignments.
The script for reproducing the analyses requires the following programs:
- phylobayes mpi, a Bayesian phylogenetic analysis program, version 1.9: https://github.com/bayesiancook/pbmpi.git
- norm cv: a program for simulating and computing model fit under a special multivariate model described in the article: https://github.com/bayesiancook/normcv.git