Abstract
dc:description.abstractThe total dependence of a virus on its host for its survival leads to a fundamental entanglement with its host’s cellular machinery. This drives a coevolutionary relationship that leaves an imprint of the host in viral genomes. The aim of this thesis was to develop machine learning approaches to identify and exploit these host predictive signals. We present methods that use these signals both to build classifiers that can assign putative information to virus genomes and to locate the discriminative features on viral proteins thereby identifying regions that are important in the host relationship. The first step aimed to identify discriminative features that capture the different aspects of the virus host relationship. We generated a range of feature sets from alternative representations of the viral genomes that each aimed to exploit the different levels of biological information present. We used a supervised machine learning approach to compare a range of feature sets for their ability to predict host taxonomic information. Next, we opened these “black box” classifiers and to extract the discriminative information learnt by the model to identify regions of a viral protein that are associated with their host relationship. We used the ‘local’ nature of some of the predictive feature sets to transform an amino acid sequence into host signals. Finally, we developed a multi-view generative mixture model, MVC, to tease apart the complex signals that are embedded in viral genomes via different evolutionary processes. This Bayesian approach uses the clustering of the data defined by labels of interest to guide the features associated with those labels into the "relevant view". The MVC model is able to identify features associated with weak effect in the data.
Degree
thesis:*- Level dc:type.qualificationlevel
- PhD
- Grantor dc:publisher.institution
- University of Glasgow
- Year dc:date.issued
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Young, Francesca
Subjects
dc:subject × 2Rights
- Language dc:language
- en
Identifiers
dc:identifier.*- DOI dc:identifier.doi
- https://doi.org/10.5525/gla.thesis.82842