Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.716049
Title: Investigating the ability of machine learning techniques to provide insight into the aetiology of complex psychiatric genetic disorders
Author: Vivian-Griffiths, Timothy
Awarding Body: Cardiff University
Current Institution: Cardiff University
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Thesis embargoed until 01 Jun 2018
Access from Institution:
Abstract:
One of the biggest challenges in psychiatric genetics is examining the effects of interactions between genetic variants on the aetiologies of complex disorders. Current techniques involve looking at linear combinations of the variants, as considering all the possible combinations of interactions is computationally unfeasible. The work in this thesis attempted to address this problem by using a machine learning model called a Support Vector Machine (SVM). These algorithms are capable of either building linear models or using kernel methods to consider the effects of interactions. The dataset used for all of the experiments was taken from a study looking into sufferers of treatment-resistant schizophrenia receiving the medication, Clozapine, with controls taken from the Wellcome Trust Case/Control Consortium study. The first experiment used information from the individual Single Nucleotide Polymorphisms (SNPs) as inputs to the SVMs, and compared the results with a technique called a polygenic score, a linear combination of the risk contributions of the SNPs that provides a single risk score for each individual. When more SNPs were entered into the models, one of the non-linear kernels provided better results than the linear SVMs. The second experiment attempted to explain this behaviour by using simulated phenotypes made from different contributions of main effects and interactions. The results strongly suggested that interactions were making a contribution. The final experiment looked at using risk scores made from gene sets. The models identified a set involved in synaptic development that has been previously implicated in schizophrenia, and when the scores from the individual genes were entered, the non-linear kernels again showed improvement, suggesting that there are interactions occurring between these genes. The conclusion was that using SVMs is an effective way to assess for the possible presence of interactions, before searching for them explicitly.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.716049  DOI: Not available
Keywords: R Medicine (General)
Share: