Improvement of sample classification and metabolite profiling in 1H-NMR by a machine learning-based modelling of signal parameters

Identificador:  TDX:2860

Handle: https://hdl.handle.net/20.500.11797/TDX2860

Autores:  Cañueto Rodríguez, Daniel

Resumen:
NMR is an analytical platform used to quantify the metabolites present in metabolomics samples. 1H-NMR spectra show multiple metabolite signals, each one with three parameters (chemical shift, half bandwidth, intensity) which can show reactivity to the sample conditions. This reactivity is a challenge for the optimization of the lineshape fitting of spectra necessary to perform the automatic metabolite profiling of samples. The aim of this PhD thesis was the exploration of the use of trending machine learning (ML)-based techniques and of robust ML-based workflows to model and then exploit the information present in the different parameters collected for each signal during the metabolite profiling of 1H-NMR datasets. In particular, the applications considered were the enhanced classification of samples in metabolomics studies and the enhancement of the quality of automatic profiling in 1H-NMR datasets. in addition to the achievement of these goals, additional achievements (e.g., the generation of a new open-source tool able to solve challenges in the profiling of complex matrices) was also fulfilled.