Skip to main content
. 2020 Sep 24;10:15652. doi: 10.1038/s41598-020-72567-6

Table 2.

Age prediction performances of the different statistical models on the training and testing sets.

Model Best performance from Training (T)/Testing (V) setsa Number of CpGs CpG combination Training set Testing set
R MAD RMSE R MAD RMSE
Zbiec-Pierkarska 1 2 CpG5,7 0.918 6.885 9.127 0.932 6.397 8.803
MQR T 9 CpG1–2 & 4–6 & CpG22,42,62–72 0.945 5.133 6.975 0.950 4.773 6.730
V 8 CpG4–6 & CpG22–42,62–72 0.941 5.229 7.184 0.953 4.574 6.559
SVMr T 6 CpG1–3,5–7 0.956 4.555 6.229 0.953 4.464 6.544
V 5 CpG2–3,5–7 0.9546 4.6139 6.3257 0.9534 4.4101 6.4919
SVMl T 7 CpG1–7 0.935 5.575 7.531 0.943 5.221 7.194
V 5 CpG2–6 0.930 5.650 7.793 0.945 5.130 7.058
SVMp T 7 CpG1–7 0.799 9.946 13.046 0.830 9.734 12.124
V 5 CpG3–7 0.778 10.456 13.582 0.833 9.465 12.098
GBR T 7 CpG1–7 0.992 1.993 2.627 0.953 4.549 6.520
V 5 CpG2,4–7 0.989 2.378 3.121 0.955 4.426 6.398
mMDA T 3 CpG1,5–6 0.933 5.650 7.625 0.940 5.320 7.357
V 3 CpG2,5–6 0.929 5.801 7.855 0.943 5.231 7.223

aFor each statistical model, both CpG combinations giving the best age prediction accuracy according to the training (T) and testing (V) sets were included in the table.