Fig. 3

Robustly well-predicted metabolites. A Top 20 robustly well-predicted metabolites. Diamonds’ centers represent the random-effects model’s estimated mean effect size (mean predictability) and diamonds’ widths represent the mean’s 95% confidence interval. The numbers in square brackets represent the number of datasets in which the metabolite was available. A star beside the number of datasets indicates that in one or more of the datasets, this metabolite was annotated with low confidence (see “Methods” section). The red dashed line represents a Spearman’s correlation of 0.3, which we defined as the threshold for a successful predictive model. B Forest plots of 3 robustly well-predicted metabolites. Yellow boxes represent predictability estimates from each dataset, with box sizes proportional to the number of samples and horizontal lines denoting the 95% confidence interval for each estimate. The diamonds and red dashed line are as above. Prediction intervals (dark red interval) further indicate the predicted range of the effect in 95% of unobserved studies [54]. C An illustration of a part of the bile metabolism pathway (including primary bile acids, secondary bile acids, and other bile components). Colored circles indicate whether the metabolite was robustly well-predicted by the gut microbiome (green), not robustly well-predicted (gray), or not included in the analysis (white). D An illustration of parts of the polyamines metabolic pathways. Color coding is as in panel C.