Boosting the Performance of Bayesian Divergence Time Estimation with the Phylogenetic Likelihood Library

摘要

We present a substantially improved and parallelized version of DPPDiv, a software tool for estimating species divergence times and lineage-specific substitution rates on a fixed tree topology. The improvement is achieved by integrating the DPPDiv code with the Phylogenetic Likelihood Library (PLL), a fast, optimized, and parallelized collection of functions for conducting likelihood computations on phylogenetic trees. We show that, integrating the PLL into a likelihoodbased application is straight-forward since it took the first author (DD) a programming effort of only one month, without having prior knowledge of DPPDiv, nor the PLL. We achieve sequential speedups that range between a factor of two to three and near-optimal parallel speedups up to 48 threads on sufficiently large datasets. Hence, with a programming effort of one month, we were able to improve DPPDiv's time-to-solution on parallel systems by two orders of magnitude and also to substantially improve its ability to infer divergence times on large-scale datasets.