2.7 | Regression analysis
The correlation between methylation rate and age for both gene regions
was examined using a single linear regression analysis. The effect of
sex and female nursing state on methylation rate in both gene regions
was assessed using analysis of covariance (ANCOVA).
To develop the age estimation model, we used the support vector
regression (SVR). We constructed three models and assessed their
precision and accuracy, as listed below:
Model 1: GRIA2 methylation rate + CDKN2A methylation rate
Model 2: GRIA2 methylation rate + CDKN2A methylation rate
+ sex
Model 3: GRIA2 methylation rate + CDKN2A methylation rate
+ female nursing state
Leave-one-out cross-validation (LOOCV) was performed to validate the
overfitting of these models. Precision and accuracy were calculated both
before and after LOOCV. All computations were performed using R (version
4.0.2) statistical software (R Core Team, 2020). The R package
“Pamesures” (Wang & Li, 2018), “e1071” (Meyer et al ., 2022),
“MuMIn” (Bartoń, 2022), and “car” (Fox & Waisberg, 2019) were used
for the analysis. The output of coefficient of “a ” was carried
out using the “nls” command. The two parameters, “cost” and
“epsilon” for the SVR models were optimized using the “tune” command
with the fixed set of “type = eps-regression, kernel = radial, gamma =
0.5”. The coefficient of determination (R2 )
and mean absolute error (MAE) was used to indicate how well an estimated
age fitted the model. Differences were considered significant atp < 0.05 for all analyses.