Priority Research Paper Volume 16, Issue 22 pp 13452—13504

Cell-type specific epigenetic clocks to quantify biological age at cell-type resolution

class="figure-viewer-img"

Figure 6. Construction and validation of hepatocyte clock. (A) Overall strategy used to train a hepatocyte specific clock from bulk liver tissue samples. Plots show how the root mean square error (RMSE) changes as a function of the penalty parameter in the elastic net regression model, and scatterplot at the bottom compares the predicted and chronological ages for the optimal model defining the hepatocyte-clock. Pearson Correlation Coefficient (PCC), Median Absolute Error (MedAE) and regression P-value are given. (B) Validation of the hepatocyte and liver clocks (latter trained on age-DMCs, not hepatocyte-DMCTs) in a primary hepatocyte culture. Also shown is the result for Horvath’s clock. PCC, MedAE and P-values from a linear regression are given. (C) Barplot comparing the PCC-values of the hepatocyte and liver clocks across several liver and non-liver tissue datasets. Boxplots compare the PCC values of the hepatocyte and liver clocks in the liver/hepatocyte dataset compared to all others. P-value is from a one-tailed Wilcoxon rank sum test. (D) Boxplots compare the extrinsic age-acceleration (EAA) of the liver, hepatocyte and Horvath clocks in two independent liver DNAm datasets, one profiling non-alcoholic fatty liver disease (NAFLD) cases and the other profiling liver-tissue from obese (BMI >35) individuals. In each case, the P-value is from a one-tailed Wilcoxon rank sum test comparing the distribution to a mean value of 0 (red dashed line). (E) Boxplots of EAA from the three clocks in an independent DNAm liver dataset, with samples stratified according to BMI levels, as shown. For each panel, we give the PCC and linear regression test P-value.