Combining NMR and LC-MS Using Hierarchical Modeling: Untargeted Metabolomics Study of Opium Users and Healthy Controls

Every year, a large number of people lose their lives due to opioid use, and a much larger number of people also suffer from opioid-related disorders. In addition to sedative effects, opioid use can have devastating effects on other parts of the body [1]. Identifying the metabolites that affect these lesions will be very valuable. Metabolic samples are very complex and their analysis requires a lot of information from the system. Both nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS) play important roles in metabolomics. Currently, the vast majority of metabolomics studies use either NMR or MS separately, and variable selection strategies that combines NMR and MS for biomarker identification and statistical modeling has not been well developed yet [2]. In this study, ultra-performance liquid chromatography (UPLC) high-resolution mass spectrometry (UPLC-HRMS) and NMR data obtained by the analysis of urine samples of Golestan Cohort opium users were fused hierarchically for identification of biochemical perturbations in opium users [3]. Here, principal component analysis (PCA) is applied separately to LC-HRMS and NMR data (lower-level stage). Then significant scores for each data are extracted from the PCA results and merged. Finally, partial least square discriminant analysis (PLS-DA) applied to fused data (higher-level stage). Consequently, the discriminating biological markers between healthy and addicted samples were uncovered. We demonstrated that variable selection is vitally important in fused NMR and HRMS data. The combined approach was more reliable than each individual modeling of NMR or LC-HRMS data with significantly improved prediction accuracy. Using this approach, a subset of metabolites (180 and 120 features from LC-HRMS and NMR data, respectively) responsible for an improved binary class separation was selected and investigated.