File:Cross-validation study of six epigenetic clocks for rat (biological aging measurement).webp

English: "A–D Four epigenetic clocks that were trained on rat tissues only. All estimates of epigenetic age and their matching chronological ages are measured in years. E–H Results for two clocks that were trained on both human and rat tissues. Leave-one-sample-out estimate of DNA methylation age (y-axis, in units of years) versus chronological age for A rat pan-tissue, B rat brain, C rat blood, and D rat liver clock. Dots are colored by A tissue type or B brain region. E and F “Human-rat” clock estimate of absolute age. G, H Human-rat clock estimate of relative age, which is the ratio of chronological age to the maximum lifespan of the respective species. Tenfold cross-validation estimates of age (y-axis, in years) in E, G human (green) and rat (orange) samples and F, H rat samples only (colored by tissue type). Each panel reports the sample size, correlation coefficient, median absolute error (MAE)" "To arrive at unbiased estimates of the six epigenetic clocks, we used (a) cross-validation of the training data and (b) evaluation with an independent test data set. The cross-validation study reports unbiased estimates of the age correlation R (defined as Pearson correlation between the age estimate (DNAm age) and chronological age) as well as the median absolute error (Fig. 1). The cross-validation estimates of the age correlations for all six clocks are higher than 0.85. The four rat clocks exhibited median absolute errors that range from 0.137 years (1.6 months) for the rat blood clock to 0.182 years (2.2 months) for the rat pan-tissue clock (Fig. 1A–D). The human-rat clock for age generated an age correlation of R = 0.99 when both species are analyzed together (Fig. 1E) but is lower when the analysis is restricted to rat tissues alone (R = 0.85, Fig. 1F). In contrast, the human-rat clock for relative age exhibits high correlation regardless of whether the analysis is done with samples from both species (R = 0.96, Fig. 1G) or only with rat samples (R = 0.94, Fig. 1H). This demonstrates that relative age circumvents the skewing that is inherent when absolute age of species with very different lifespans are measured using a single formula. This is due in part to the unequal distribution of training data at the opposite ends of the age range."