Fig. 2. The age predictor models’ sample distribution and prediction accuracy.

a Histogram showing age distribution of all samples used in Mitra clocks development, split into training (blue) and independent (orange) sets. b Age distribution of the independent dataset stratified by data cohorts included. c Age distribution of the training dataset stratified by data cohorts included. Cohorts in (b, c) represent separate sampling and sequencing batches. d, e Predicted versus chronological age in the independent test set for the single-CpG model, MitraSolo (R² = 0.88, MAE = 4.09) (d) and the region-based model, MitraCluster (R² = 0.89, MAE = 4.00) (e), demonstrating high accuracy in both models. The independent set samples cover, on average, 98.1% of the MitraSolo CpGs and 97.9% of the MitraCluster CpGs.