Fig. 6

Predicted vs. measured concentrations of inorganic ions and some metabolites in real urine samples. Plots of the predicted vs. measured concentrations of 11 inorganic ions in 60 real urine samples, and of 6 among the most active metabolites in 120 real urine samples. The panels show: a chloride (Cl⁻), b sodium (Na⁺), c calcium (Ca²⁺), d potassium (K⁺), e magnesium (Mg²⁺), f phosphate (PO₄ ³⁻), g sulfate (SO₄ ²⁻), h lithium (Li⁺), i rubidium (Rb⁺), j zinc (Zn²⁺), k aluminum (Al³⁺), l creatinine, m dimethylamine, n creatine, o hippurate, p guanidoacetate, and q tartrate. Additional panels for another set of six metabolites are reported in Supplementary Fig. 5. The concentrations of the inorganic ions were determined in clinical analysis and analytical chemistry laboratories, and the metabolite concentrations were estimated by NMR lineshape fitting analyses (Methods). Horizontal error bars for ions indicate confidence intervals according to the estimated errors of the analytical technique used (Methods), and the vertical ones both for ions and active metabolites are based upon the accuracy of each predictive model, extracted from the relative root mean square error (rRMSE) values. In turn, the rRMSE values were calculated by the predictive efficiency of the models in the 40 randomly prepared artificial urine mixtures (Supplementary Fig. 3) and the 4501 artificial urine spectra. For the cases of h Li⁺ and i Rb⁺, 10 and 6 urine samples, respectively, lie outside the normal concentration range in urine, as highlighted by an additional panel. It can be seen that the models are able to predict also these outliers with high accuracy. In addition, the diseases involving alterations of each of the reported ions are reported in orange panels, with the black arrows in the parentheses showing the concentration trend of the ion for each disease