Fig. 4

Partial correlation networks of the discriminatory metabolites between treatment groups, applying untargeted and targeted metabolomics. Partial correlation networks including as independent variables the metabolite changes identified as significantly different between treatment groups in multivariate analysis and BMI change, using untargeted (a) and targeted metabolomics (b). The networks were generated for each treatment group and were estimated based on a graphical LASSO algorithm, with extended Bayesian information criterion (EBIC) to select the model complexity. Nodes are measured variables and edges (lines) are inferred associations (width: strength; color: sign). Green and red edges represent positive and negative correlation, respectively. In networks corresponding to untargeted metabolite changes (a), features corresponding to the same metabolite are clustered and encircled in colored spheres with metabolites named in matching color in the legend