Building Better Trees: Multidimensional optimization algorithms improves accuracy of maximum likelihood phylogenetic inference

In order to understand how a biological system functions, it is often useful to understand how that system came to be, or evolved. Modern evolutionary biology relies on building trees (called phylogenies) from molecular sequence data using models of molecular evolution. These models include many free parameters that must be optimized. Virtually everyone optimizes their phylogeny using Unimax, but here we show that Multimax is more accurate—-and more computationally demanding. I implemented a Multimax approach to phylogenetic optimization, using Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm. Here I show that Multimax significantly improves the accuracy of ML phylogenetic inference, especially for large trees. Although Multimax is more accurate than Unimax, Multimax introduces significant computational costs. Our results strongly suggest that Unimax should be avoided for phylogenetic inference, but there is no easy panacea for phylogenetic optimization.