Skip to main content

Table 1 Age profile and test set prediction performance for cohorts used in cAge predictor training and testing. Predictions were made using a LOCO approach, where each cohort was excluded in training and the resulting model was used for testing (see Methods). Models were trained on age, and if an individual was predicted to be under 20, their prediction was re-estimated considering models trained on log(age). External cohort information taken from Zhang et al. [5]. r column states Pearson correlation, RMSE the root mean squared error, and MAE the median absolute error

From: Refining epigenetic prediction of chronological and biological age

      

Prediction accuracy

Cohort

N

Mean age (SD)

Age range

NFemales (%)

Tissue

r

RMSE

MAE

GS

18,413

47.5 (14.9)

[17.1, 98.5]

10,833 (58.8%)

Blood

-

-

-

LBC192120,21

692

82.3 (4.3)

[77.8,90.6]

401 (57.9%)

Blood

0.659

4.050

2.466

LBC193620,21

2796

73.6 (3.7)

[67.7,80.9]

1356 (48.5%)

Blood

0.685

3.311

2.099

GSE7277522

335

70.2 (10.3)

[36.5, 90.5]

138 (41.2%)

Blood

0.949

3.275

1.843

GSE7887422

259

68.8 (9.7)

[36.0, 88.0]

113 (43.6%)

Saliva

0.875

6.826

4.333

GSE7277322

310

65.6 (13.9)

[35.1, 91.9]

150 (48.4%)

Blood

0.945

4.611

2.068

GSE7277722

46

14.7 (10.4)

[2.2, 35.0]

31 (67.4%)

Blood

0.942

4.211

2.505

GSE41169a,23

95

31.6 (10.3)

[18.0, 65.0]

28 (29.5%)

Blood

0.975

2.869

1.947

GSE402794

656

64.0 (14.7)

[19.0, 101.0]

338 (51.5%)

Blood

0.969

3.697

2.074

GSE42861a,24

689

51.9 (11.8)

[18.0, 70.0]

492 (71.4%)

Blood

0.972

4.498

3.563

GSE53740a,25

383

67.8 (9.6)

[34.0, 93.0]

155 (40.5%)

Blood

0.921

4.443

2.797

  1. aSome cohorts contain case/control data. GSE41169: schizophrenia 62, control 33; GSE42861: rheumatoid arthritis 354, control 335; GSE53740: Alzheimer’s disease 15, corticobasal degeneration 1, frontotemporal dementia (FTD) 121, FTD/MND 7, progressive supranuclear palsy 43, control 193, unknown 4