Use this URL to cite or link to this record in EThOS:
Title: Multi-layer perceptron artificial neural network predictive modelling of genomic and mass spectrometry data in bioinformatics
Author: Lancashire, Lee James
ISNI:       0000 0001 3604 5364
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 2006
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
The development of proteomic and genomic applications for the research into different diseases has paved the way for the development of novel approaches for the way in which these systems can be investigated. This provides a novel insight into how proteins and genes are being regulated under different conditions. These approaches are self-limited by the volume of data which they produce, with the majority often being noisy and redundant. Therefore these technologies must be coupled with appropriate computational approaches that are capable of identifying components that are the most important in differentiating between disease states of interest. These must be robust enough to cope with data of this size and nature in order to provide an in depth understanding of these complex proteomic and genomic patterns. This in turn will lead to methods for prognosis and diagnosis of diseases such as cancer, by providing an insight into the proteins and genes which are being expressed differentially depending on the current status of the disease. The research contained within this thesis describes the development and validation of multi-layer perceptron Artificial Neural Network based methodologies for variable selection, biomarker identification and predictive modelling of mass spectrometry and genomic data. Many datasets were used from a range of different sources, such as the mass spectrometry analysis of bacterial pathogens, the mass spectrometry analysis of patients suffering from different grades of melanoma, and gene expression analysis of patients with breast cancer. Results showed that robust and reproducible predictive models could be generated, which predicted class to extremely high accuracies (greater than 95 %) for blind datasets. These approaches were enhanced further to allow for the interrogation of biomarkers identified during the course of the analyses with techniques such as response surface analysis and population structure analysis. Response surfaces showed the direction of response of a biomarker of interest, in relation to whether it was being up or down regulated in a given disease outcome under study. As an adjunct population profiling showed the potential for identifying sub-groups of patients which could subsequently be used to identify those at risk of disease spread based upon their genetic profiles. Finally methods for the derivation of gene regulatory networks have been proposed which allows interactions and pathways to be derived to show how the change in expression of one gene causes a resulting change in many others. As such the results from the experimental work performed in this thesis have resulted in novel contributions to the field of bioinformatics.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available