Recoupled-STOCSY-based co-expression network analysis to extract phenotype-driven metabolite modules in NMR-based metabolomics dataset.

Nuclear magnetic resonance (NMR)-based metabolomics study usually involves spectral preprocessing, identification of biomarkers and interpretation of biological processes and pathogenesis, however, the traditional procedure is bound to inborn defects. In this study, a new analytical frame was proposed to assist spectral alignment and dimensionality reduction, screen the differential metabolites and get biological explanation of the metabolic network by combing weighted gene co-expression network analysis (WGCNA) and recoupled statistical total correlation spectroscopy (RSTOCSY). The performance of RSTOCSY-based WGCNA method was evaluated by the NMR dataset of serum from coronary heart disease with diabetes mellitus (CHDDM) patients. The statistical recoupling of variables (SRV) was successfully used to categorize the whole dataset into a number of superclusters of signals and served to spectral alignment, and its effectiveness was confirmed by the wine dataset with a larger spectral drift. Three phenotype-driven metabolite modules related to CHDDM were identified from the dataset by WGCNA, and 22 metabolites were further identified from the three modules according to the metabolic correlations within or between modules, and 40 significant metabolic correlations were observed from the intra- and inter-metabolites in the 2D pseudospectrum. These modules involve amino acid metabolism, microbial metabolism and glucose metabolism, and their analysis of metabolite network diffusion revealed a new discovery that the ferroptosis pathway is related to CHDDM. This RSTOCSY-based WGCNA approach provides an effective analysis workflow for information recovery and structure identification of metabolites and improving interpretability and understanding of the disease pathogenesis.