Table 1.
The midlife ARIC and published PACs; Visit 2 (N = 2,993 in training set, N = 1,496 in test set) | ||
---|---|---|
| ||
midlife ARIC PAC | midlife Lehallier’s PAC | |
| ||
Number of aptamers in PAC | 788 | 415 |
Hyperparameter value (lambda) | 0.11 | 1.03 |
Correlation in the training seta | 0.92 | 0.82 |
Correlation in the test seta | 0.80 | 0.76 |
MAE in the training seta | 1.50 | 2.13 |
MAE in the test seta | 2.19 | 2.39 |
| ||
The late-life ARIC and published PACs; Visit 5 (N = 630 in training set, N = 315 in test set) | ||
| ||
late-life ARIC PAC | late-life Lehallier’s PAC | |
| ||
Number of aptamers in PAC | 135 | 415 |
Hyperparameter value (lambda) | 0.46 | 4.42 |
Correlation in the training seta | 0.84 | 0.84 |
Correlation in the test seta | 0.71 | 0.63 |
MAE in the training seta | 1.47 | 1.71 |
MAE in the test seta | 2.36 | 2.37 |
Abbreviations: MAE - median absolute error; PAC - proteomic ageing clock.
Among healthy participants at Visit 2 and Visit 5, we randomly selected two-thirds of healthy participants at each visit and used them as the training set at the corresponding visit; the remaining one-third of healthy participants at each visit was used as the test set at the corresponding visit.