Table 6. Accuracy of meta-feature signatures involving differences in analyte abundances in classifying controls and MCI progressors.
| Signature | Cross-Validation | Training Set | Test Set | ||||||
| Sens | Spec | MCC | Sens | Spec | MCC | Sens | Spec | MCC | |
| 8-metafeature signature with APOE | |||||||||
| a Full set of samples | 94.7 | 78.1 | 0.74 | 99.1 | 96.3 | 0.96 | 93.2 | 64.8 | 0.61 |
| b Size-matched groups | 90.2 | 87.2 | 0.77 | 98.9 | 98.1 | 0.97 | 85.6 | 86.7 | 0.73 |
| 7-metafeature signature without APOE | |||||||||
| Full set of samples | 95.6 | 78.1 | 0.76 | 97.9 | 95.9 | 0.93 | 96.0 | 72.6 | 0.73 |
| Size-matched groups | 83.3 | 87.6 | 0.71 | 90.7 | 99.2 | 0.91 | 90.7 | 83.0 | 0.74 |
a
The full set of samples contained data on 54 Controls and 163 MCI Progressors.
b
The size-matched groups contained data on 54 Controls and 54 MCI Progressors. For each of these datasets, all samples were used for cross-validation, whereas training and test sets were created by dividing datasets into two equal subsets.