Reconstruction of Metabolic Association Networks Using High-throughput Mass Spectrometry Data

摘要

Graphical Gaussian model (GGM) has been widely used in genomics and proteomics to infer biological association networks, but the relative performances of various GGM-based methods are still unclear in metabolomics. The association between two nodes of GGM is calculated by partial correlation as a measure of conditional independence. To estimate the partial correlations with small sample size and large variables, two approaches have been introduced, which are arithmetic mean-based and geometric mean-based methods. In this study, we investigated the effects of these two approaches on constructing association metabolite networks and then compared their performances using partial least squares regression and principal component regression along with shrinkage covariance estimate as a reference. These approaches then are applied to simulated data and real metabolomics data.