Soutenance de thèse de M.Viet Dung DUONG du CRMN sous la direction de M. Torsten HERRMANN
Nuclear Magnetic Resonance (NMR) has become one of the most powerful and versatile spectroscopic techniques in analytical chemistry with applications in many disciplines of scientific research. A downside of NMR is however the laborious data analysis workflow that involves many manual interventions. Interactive data analysis impedes not only on efficiency and objectivity, but also keeps many NMR application fields closed for non-experts. Thus, there is a high demand for the development of unsupervised computational methods. This thesis introduces such unattended approaches in the fields of metabonomics and structural biology.
A foremost challenge to NMR metabolomics is the identification of all molecules present in complex metabolite mixtures that is vital for the subsequent biological interpretation. In this first part of the thesis, a novel numerical method is proposed for the analysis of two-dimensional HSQC and TOCSY spectra that yields automated metabolite identification. Proof-of principle was successfully obtained by evaluating performance characteristics on synthetic data, and on real-world applications of human urine samples, exhibiting high data complexity.
NMR is one of the leading experimental techniques in structural biology. However the conventional process of structure elucidation is quite elaborated. In this second part of the thesis, a novel computational approach is presented to solve the problem of NMR structure determination without explicit resonance assignment based on three-dimensional TOCSY and NOESY spectra. Proof-of principle was successfully obtained by applying the method to an experimental data set of a 12-kilodalton medium-sized protein.
CRMN-ISA  Amphithéâtre : La Doua