Probabilistic modelling of liquid chromatography time-of-flight mass spectrometry

Liquid Chromatography Time-of-Flight Mass Spectrometry (LC-TOFMS) is an analytical platform that is widely used in the study of biological mixtures in the rapidly growing fields of proteomics and metabolomics. The development of statistical methods for the analysis of the very large data-sets that are typically produced in LC-TOFMS experiments is a very active area of research. However, the theoretical basis on which these methods are built is currently rather thin and as a result, inferences regarding the samples analysed are generally drawn in a somewhat qualitative fashion. This thesis concerns the development of a statistical formalism that can be used to describe and analyse the data produced in an LC-TOFMS experiment. This is done through the derivation of a number of probability distributions, each corresponding to a different level of approximation of the distribution of the empirically obtained data. Using such probabilistic models, statistically rigorous methods are developed and validated which are designed to address some of the central problems encountered in the practical analysis of LC-TOFMS data, most notably those related to the identification of unknown metabolites. Unlike most existing bioinformatics techniques, this work aims for rigour rather than generality. Consequently the methods developed are closely tailored to a particular type of TOF mass spectrometer, although they do carry over to other TOF instruments, albeit with important restrictions. And while the algorithms presented may constitute useful analytical tools for the mass spectrometers to which they can be applied, the broader implications of the general methodological approach that is taken are also of central importance. In particular, it is arguable that the main value of this work lies in its role as a proof-of-concept that detailed probabilistic modelling of TOFMS data is possible and can be used in practice to address important data analytical problems in a statistically rigorous manner.EThOS - Electronic Theses Online ServiceWellcome TrustGBUnited Kingdo