Fig. 2
From: Predictive metabolomic profiling of microbial communities using amplicon or metagenomic sequences
MelonnPan accurately predicts metabolite features based on metagenomic sequence profiles. a From a panel of 466 metabolites whose identities were confirmed against laboratory standards, these 50 were the best predicted by MelonnPan with unique Human Metabolome Database (HMDB) identifiers (as measured by Spearman correlation (r) of predicted versus measured profiles in the Netherlands inflammatory bowel disease (NLIBD) independent validation cohort). All metabolites shown have r > 0.3 over a total n = 65 NLIBD samples. Two representative metabolites are shown in red. b Measured and predicted metabolite profiles across a single representative sample were strongly and significantly associated across 107 labelled metabolites (including both unique and non-unique compounds that were well predicted by MelonnPan). c Representative significant prediction for cholestenone in the test set. d Representative statistically significant prediction for pantothenate in the NLIBD validation data (values are relative abundances). See Supplementary Figs. 1–2 and Supplementary Data 1–6 for full results. For each scatter plot, the best fitting regression line is also shown (in red)
