Dissertation Defense Details
Error and Uncertainty in Computational Phylogenetics
| Author: | Victor Hanson-Smith |
|---|---|
| Date: | November 09, 2011 |
| Time: | 14:00 |
| Location: | 220 Deschutes |
| Committee: | John Conery (Chair) Sarah Douglas Daniel Lowd Joe Thornton |
Abstract
Over the last two centuries of human thought, the discovery of biological evolution profoundly changed -- and continues to change -- our perception of the living world. Evolutionary history can be difficult to study because evolution operates on timescales that are much longer than the length of a human lifetime. The techniques of computational phylogenetic ancestral reconstruction tackle this problem by inferring evolutionary history from contemporary genomes, using probabilistic Markov models of molecular sequence evolution. Optimizing these models is NP-hard, so clever search heuristics are required. In this dissertation, I show how we can increase the accuracy of phylogenetic inference by using an alternative heuristic based on conjugate-gradient methods. I further show that the reconstruction of ancestral molecular sequences is robust to statistical uncertainty about the underlying evolutionary tree. Finally, I demonstrate the use of phylogenetic ancestral reconstruction within an analysis pipeline -- combining computational and wet-lab techniques -- to reconstruct the functions of an 800 million year-old molecular machine in the ancient ancestor of fungus and animals.