Fig. 4
Flowchart of the shift and concentration prediction. The flowchart of the algorithm is highlighted in purple. The algorithm starts a upon the decision of the user about the automated or manual identification of the five navigator signals: 2 doublets of doublets (dd) of citrate, 2 singlets (s) of creatinine, and 1 singlet (s) of glycine. For citrate and creatinine specific criteria are used (e.g., the J-coupling constants of the doublets as well as the peak intensity ratios), applied to all peaks detected in a scanned spectral area where each spin system is expected to resonate (Supplementary Table 1). Since no constraints can be applied for the detection of the singlet of glycine, an extra prediction model is employed that predicts the chemical shift (δ) of the glycine singlet based upon the δ of creatinine and citrate. The prediction ability of this model was tested in 1600 different urine samples and resulted in no mis-assignments of the glycine NMR signals17,18,19. Using as input the five navigator signals, the algorithm initially employs the reduced models (based on the five navigator signals) to estimate the metabolites/ions concentrations, pH, T, and then predict the δ values for the other metabolites. Afterwards, the algorithm firstly predicts the δ values, and then concentrations, pH, and T are estimated from the former. The two routes result in two predicted δ values for each signal (i.e., δ reduced and δ full), from which, according to specific selection criteria, the best results are chosen (more details in the Methods section). b The output of the algorithm consists of the 94 1H NMR signals of 63 metabolites and c of a figure for each urine NMR spectrum, where the predicted δ of each signal along with its multiplicity is reported. d Moreover, one extra figure is exported comparing the predicted concentrations of 11 ions and albumin with the normal clinical range for male/female adults (based upon Mayo Clinic, www.MedlinePlus.gov and the reference guides of the laboratories that performed the clinical analyses in the present study). Two extra files contain the predicted concentrations of 11 ions and albumin, pH, and T values, and the estimated concentration values of 40 metabolites, respectively
