Table 2.
Comparison of model performances of PLS with sevenfold cross-validation among models built on metabolites markers (M), clinical markers (C) or combination of metabolite and clinical markers (C + M).
| Dataset | Biomarker panel | nComp | R2Y | Q2 | PCV-ANOVA | RMSEE | RMSEcv | r |
|---|---|---|---|---|---|---|---|---|
| Pancreatic fat | M | 3 | 0.81 | 0.69 | 2.12E–12 | 0.72 | 0.89 | 0.90 |
| C | 1 | 0.51 | 0.46 | 1.15E–08 | 1.14 | 1.18 | 0.71 | |
| C + M | 3 | 0.81 | 0.67 | 1.12E–11 | 0.72 | 0.92 | 0.90 | |
| Liver fat | M | 3 | 0.80 | 0.66 | 6.11E–12 | 3.84 | 4.65 | 0.89 |
| C | 1 | 0.48 | 0.40 | 3.51E–07 | 5.98 | 6.37 | 0.70 | |
| C + M | 2 | 0.74 | 0.63 | 5.49E–12 | 4.25 | 4.94 | 0.86 | |
| VAT/SAT | M | 2 | 0.70 | 0.62 | 1.88E–12 | 0.13 | 0.14 | 0.83 |
| C | 2 | 0.56 | 0.44 | 1.42E–07 | 0.16 | 0.17 | 0.74 | |
| C + M | 2 | 0.74 | 0.67 | 3.98E-14 | 0.12 | 0.13 | 0.86 |
nComp: number of components; RMSEE: roo mean square error of estimation; RMSEcv: root mean square error of cross-validation; r: Pearson correlation coefficient.