Use this URL to cite or link to this record in EThOS:
Title: Development and application of chemometric methods for modelling metabolic spectral profiles
Author: Fonville, Judith Marlou
ISNI:       0000 0004 2704 5938
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
Access from Institution:
The interpretation of metabolic information is crucial to understanding the functioning of a biological system. Latent information about the metabolic state of a sample can be acquired using analytical chemistry methods, which generate spectroscopic profiles. Thus, nuclear magnetic resonance spectroscopy and mass spectrometry techniques can be employed to generate vast amounts of highly complex data on the metabolic content of biofluids and tissue, and this thesis discusses ways to process, analyse and interpret these data successfully. The evaluation of J -resolved spectroscopy in magnetic resonance profiling and the statistical techniques required to extract maximum information from the projections of these spectra are studied. In particular, data processing is evaluated, and correlation and regression methods are investigated with respect to enhanced model interpretation and biomarker identification. Additionally, it is shown that non-linearities in metabonomic data can be effectively modelled with kernel-based orthogonal partial least squares, for which an automated optimisation of the kernel parameter with nested cross-validation is implemented. The interpretation of orthogonal variation and predictive ability enabled by this approach are demonstrated in regression and classification models for applications in toxicology and parasitology. Finally, the vast amount of data generated with mass spectrometry imaging is investigated in terms of data processing, and the benefits of applying multivariate techniques to these data are illustrated, especially in terms of interpretation and visualisation using colour-coding of images. The advantages of methods such as principal component analysis, self-organising maps and manifold learning over univariate analysis are highlighted. This body of work therefore demonstrates new means of increasing the amount of biochemical information that can be obtained from a given set of samples in biological applications using spectral profiling. Various analytical and statistical methods are investigated and illustrated with applications drawn from diverse biomedical areas.
Supervisor: Holmes, Elaine ; Nicholson, Jeremy Sponsor: Analytical Chemistry Trust Fund of the Royal Society of Chemistry
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral