Fig. 3.

Associations of metabolomic data with serum ferritin. Permutation tests for the goodness-of-fit (R²Y) and goodness of prediction (Q²Y) obtained from the O-PLS model between serum ferritin and a the serum (n = 48) and e urine metabolome (n = 47) in the discovery cohort, and b the serum (n = 328) and f urine metabolome (n = 322) in the replication cohort. Significant c, d serum and g, h urine metabolites associated with serum ferritin after further validation of O-PLS identified metabolites by pSC adjusting for age, sex, BMI and country. i O2-PLS scores for the joint variation between plasma and urine metabolites and microbial families associated with serum ferritin. A model with 2 predictive components, and 0 and 1 orthogonal component for the metabolites and bacterial families blocks, was constructed based on 7-fold cross-validation. j O2-PLS joint loadings plots, where pcorr represents the correlation-scaled loadings from the gene block and qcorr represents the correlation-scaled loadings from the bacterial families block. k Heatmap for the pSC adjusted by age, BMI and country between ferritin-associated urine and l plasma metabolites with ferritin-associated bacterial families (n = 56). Only significant associations (p < 0.05) are displayed. Significant associations after a pFDR correction (pFDR < 0.05) are highlighted with a black box