Fig. 1. Metabolomic signatures associated with mortality.
(A) Volcano plot of 144 measurable metabolites. Top significant metabolites between NS and S are above the dashed line (PFDR < 0.01). Yellow and blue circles correspond to elevated and reduced levels in NS compared to S, respectively. (B) Frequency of metabolite selection by the elastic net multivariate analysis. Bars correspond to the percentage of the respective metabolite being selected by the model out of the 200 bootstrap samples. Influential metabolites are those selected >70% (dashed line) of times, with bar color denoting mean concentration higher (yellow) or lower (blue) in NS compared to S. (C) Score plot of individuals (NS in yellow; S in blue) clustered by multilevel partial least square discriminant analysis (PLS-DA) of 32 differential metabolites. The interface between the white and the gray shaded area represents the classification decision line. The first two components explained 42% of the total variance. Model performance and validity measures were as follows: cross-validated AUC = 0.96 ± 0.006, misclassification rate = 0.12 ± 0.01, DR2 = 0.74 ± 0.001 and DQ2 = 0.63 ± 0.02. (D) Correlations between the 32 differential metabolites and PLS components. Arrows denote the direction and magnitude of correlations. See fig. S2 for boxplots of concentrations of the top contributing metabolites (correlation strength > 0.5). HVA, homovanillic acid; 5-HIAA, 5-hydroxyindole acetate; p-HPA, p-hydroxyphenylacetate; C2, acetylcarnitine.