Extended Data Fig. 3: Sensitivity analyses for Fig. 3.

a, Sensitivity analysis on the effect of different analysed timespans (a proxy for lifespans) on estimates of \(N/s\). For each estimate, the human cohort was trimmed to the stated maximum age after adulthood. The top 100 age-related sites (Spearman’s rank correlation coefficient) were then recalculated on this restricted dataset, and the model was then fit. Each estimate was calculated from a sample of 150 participants (approximately the maximum sample size in the smallest age range), uniformly sampled within the given age range. Estimates displayed as mean of the posterior, with error bars indicating the 95% credible interval (log scale). Included in red is the regression line we find from our scaling analysis, i.e., what we would expect if our results were due solely to the fact that animals of longer lifespan are generally observed over longer timespans. b, As in a), except with sites selected based on variance change (White test) rather than mean change. c, Sensitivity analysis on the effect of site selection on the scaling of \(N/s\) with maximum lifespan across mammals. Each inference based on the top 500 sites that change in variance (White test) instead of mean change, and using the maximum sample size available for each mammal (minimum 110, capped at 1000 for humans). d, Example in mice of a CpG fit using a single \(N/s\) value across all mammals. e, The same data shown in d, fit using an \(N/s\) value that scales with lifespan across mammals. f, The same data shown in d, fit using the \(N/s\) values determined independently for each mammal. g, Example in humans of a CpG fit using a single \(N/s\) value across all mammals. h, The same data shown in g), fit using an \(N/s\) value that scales with lifespan across mammals. i, The same data shown in g), fit using the \(N/s\) values determined independently for each mammal.

Source Data