Related resources
Search for item elsewhere
University researcher(s)
Academic department(s)
Model and Algorithm Development in Computational Phylogenetics
[Thesis]. Manchester, UK: The University of Manchester; 2011.
Access to files
-
FULL-TEXT.PDF (pdf)
Abstract
Phylogenetic trees play an important role in many areas of biological research fromhelping us to better understand the process of evolution to drug discovery. Researchersin these areas must be confident that the methods available to them will provide a ’good’phylogenetic tree; one that adequately describes the evolution of the taxa in the tree.One use of such a good tree is in modelling gene family evolution. Gene family size isclosely related to copy number variation, with the later shown to be related to importantphenotypic effects such as causing disease or conferring drug resistance. It is thereforeimportant to establish good methods for inferring gene family evolution as a betterunderstanding of how gene families evolve could help answer many important biologicalquestions.In this thesis I investigate methods of searching for a ‘good’ phylogenetic tree and thespace of trees that these methods search. I study methods for finding the neighbours ofa tree, an important step in many tree search algorithms. I explore the structure of suchmethods and investigate how likely they are to find a ‘good’ tree. I look for patterns inthe structure of tree-space that may make tree search easier or allow the best focusing oflimited resources when searching for a ‘good’ tree.I also study methods for inferring gene family evolution. I first compare two differentmethodsforinferringgenefamilyhistory, maximumparsimonyandmaximumlikelihood.I then investigate the models used in the maximum likelihood method to determine whichmodels best describe the data and what we can learn about gene family evolution fromthose models.I conclude that the Nearest Neighbour Interchange (NNI) technique should not beused as it regularly finds ‘bad’ optima and the structure of the method means that treesearch algorithms are likely to get stuck at poor optima. Instead I recommend the useof the Subtree Pruning and Re-grafting (SPR), or other more complex, methods. I alsofind several properties of tree space that may be helpful to those designing, or using, thealgorithms. My results suggest that maximum likelihood should normally be used formodelling gene family evolution but that further research is needed into the models usedif we are to be confident in the conclusions we draw.