Metabolomics is a cutting-edge area that has been widely applied in various areas, especially in human disease mechanisms and potential biomarker studies. However, data reliability is one of the most important factors for the future applications of metabolomics, and nuclear magnetic resonance (NMR) has a great advantage in data reproducibility comparing to other platforms. While two-Dimensional NMR has been widely applied in metabolomics, the data interpretation methods for 2D data are limited. The high-resolution 1D NMR metabolites chemical shift range has been applied as an effective approach to filter 2D NMR data, however, the noise suppression is still challenging. In this study, machine learning approaches principal component regression (PCR) and partial least squares regression (PLS-R) were applied to the 2D NMR spectrum. The study was demonstrated by a mixture of standard metabolites with 2D non-uniform sampling (NUS) total correlation spectroscopy (TOSCY) data and the 1D NMR peak intensity was applied as the Y block data which showed effective noise suppression. The approach has been further tested using metabolomics samples including tea extracts and sea anemones extracts. Different data scaling methods such as auto-scaling and Pareto scaling were also tested for the performance evaluation. In summary, our study provided a new way to analyze 2D NMR data which can potentially be applied to future automatic NMR metabolite identification and interpretation in the metabolomics area.

Machine learning approaches for noise reduction in two-dimensional NMR data interpretation

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.