Automated processing of untargeted 1H-NMR metabolomics data of urine from treated TB patients
Abstract
Metabolomics is becoming an increasingly popular field of study. An analytical platform commonly used for metabolomics studies is proton nuclear magnetic resonance (1H-NMR) spectroscopy. The main challenge of 1H-NMR is that the interpretation of the spectral output is time-consuming and somewhat difficult, requiring not only a trained analyst, but one with experience. For this reason, the automation of 1H-NMR metabolomics data processing is gaining attention, leading to the invention of various software tools/applications. One such software tool is BAYESIL, a fully automated and quantitative tool focussing on 1H-NMR. However, BAYESIL has not yet been rig-orously tested to ascertain robustness and accuracy in urine. This software tool is the focus of this study, where it was compared to the typical method currently used at the Centre for Human Metabolomics (CHM) at the North-West University (NWU). The aim of this study was to determine whether the automation of the processing of untargeted 1H-NMR metabolomics data has the ca-pability to obtain a comprehensive and comparable metabolic profile of treated and untreated tuberculosis (TB) patient urine samples which have been specifically selected for this study due to their complexity.
TB urine samples were analysed by a 1H-NMR spectrometer, along with healthy control samples. The data matrices containing the results of each method were subjected to the same statistical analysis via the online tool, MetaboAnalyst, to identify important metabolites and obtain a metab-olite profile of the samples. While working with MetaboAnalyst, a few areas were identified where a novice user easily makes mistakes. Tips on how to navigate around these areas were included in Chapter 4 as guideposts to help future novices along the way. The metabolite profiles obtained by each method were then compared. The main findings of this study were that the automated BAYESIL method seems to struggle with the identification of some metabolites (such as those arising from TB disease or medication) due to its limited library. BAYESIL also had some difficulty with the coverage of the spectra and identified metabolites with a lower confidence and lower specificity. On the other hand, the manual method was able to comfortably identify metabolites relating to TB medication and identified metabolites with higher specificity and higher confidence.
These results indicate that the BAYESIL method is currently better suited for less detailed re-search, such as shotgun metabolic profiling. If a more detailed representation of the metabolism is desired, it is recommended to use the manual method or even combine these methods.