A Scalable Implementation of Complex Substitution Models for Phylogenetic Inference from Large-Scale Genomic Data


Prof. Dr. Gert Wörheide
Lehrstuhl für Paläontologie & Geobiologie
LMU München

Project Summary

Mixture models in Bayesian phylogenetic analyses have proven to be useful in reconstructing deep-time phylogenies and reducing model artifacts like long-branch attraction; however, this is at the cost of computational and model complexity. In this project, we have added finite Bayesian mixture models to the software RevBayes, and have resolved the identifiability problem for mixture profile parameters by imposing and preserving a partial ordering during MCMC analysis. In addition, we have integrated the phylogenetic likelihood software, BEAGLE, into RevBayes to improve performance by utilizing vectorization, threading,  and GPU processors. Preliminary testing has shown a 2-3x speedup for a fixed number of iterations when using the CPU.