| dc.description.abstract | The inference of species divergence time is a key step in the study of phylogenetics. Methods have been available for the last ten years to perform the inference, but, there are two significant problems with these methods. First, the performance of the methods does not yet scale well to studies with hundreds of taxa and thousands of DNA base pairs. A study of 349 taxa was estimated to require over 9 months of processing time. Second, the accuracy of the inference process is subject to bias and variance in the specification of model parameters that is not completely understood. These parameters include both the topology of the phylogenetic tree and, more importantly for our purposes, the set of fossils used to calibrate the tree.
In this work, we present new algorithms and methods to improve the performance of the divergence time process. We demonstrate a new algorithm for the computation of phylogenetic likelihood and experimentally illustrate a 90% improvement in likelihood computation time on the aforementioned dataset of 349 taxa with over 60,000 DNA base pairs. Additionally we show a new algorithm for the computation of the Bayesian prior on node ages that is experimentally shown to reduce the time for this computation on the 349 taxa dataset by 99%.
Using our high performance methods, we present a novel new method for assessing the level of support for the ages inferred. This method utilizes a statistical jackknifing technique on the set of fossil calibrations producing a support value similar to the bootstrap used in phylogenetic inference.
Finally, we present efficient methods for divergence time inference on sets of trees based on our development of subtree sharing models. We show a 60% improvement in processing times on a dataset of 567 taxa with over 10,000 DNA base pairs. | en |
