Novel chemometric and computational methods for modelling cardiovascular metabolic phenotypic data

Abstract

Metabolic phenotyping, the analysis of the chemical content of human tissue and biofluid samples with either NMR or MS profiling, holds great potential for diagnostics, patient stratification and as a molecular epidemiology approach. But for this potential to be fulfilled, adequate bioinformatic and computational methodologies are required to help extract information from the vast amounts of complex data generated by the analytical platforms, whose high-throughput capacity, phenotypic coverage and cost-effectiveness are ever increasing. The work described in this thesis strives to contribute to this necessary development of bioinformatic work-flows and data analysis methods with the aim of assisting in the processing, analysis, interpretation and design of metabolic phenotyping studies and approaches. Special emphasis was given to the development of bioinformatic solutions for the pre- processing of very high-throughput direct infusion mass spectrometry assays, the adaptation of curve-fitting workflows for the automated annotation and quantification of metabolites from the proton NMR spectra of human serum and plasma biofluid, and to the development of computational and simulation methodologies for the design of metabolic phenotyping studies. The main outcomes of the work described in this thesis were the development of a python open source software for efficient handling and analysis of large direct infusion mass spectrometry datasets, the demonstration of the feasibility and the advantage of applying BATMAN, an automated curve fitting open source software for the routine analysis of NMR spectra of human blood products, and finally, the development of flexible methodologies for sample size determination, power analysis and design of metabolic phenotyping studies.