Extended Data Fig. 7: Power to detect a genetic effect on lifespan using different study designs.

(A) Y-axis represents the proportion of simulated true positive sites (with a significant change in allele frequency across the lifespan) that are detected at p < 10⁻⁶. Shown are results from simulations that mimic the “fly” design used here, where individuals are sampled at the beginning and end of their life, versus the “human” design of Mostafavi et al.¹⁴, where individuals were only genotyped in middle and old age. We performed simulations for the fly and human designs where true genotypes were estimated without error, as well as simulations where genotypes were derived from a low coverage (~1x) sequencing approach and therefore estimated with some error. The curves show simulation results for starting minor allele frequencies from 0.05-0.4, and effect sizes were always simulated as a 20% change in allele frequency from the youngest to the oldest age bin. Further details on the simulations and analysis of simulated data are provided in Text S11. (B) The number of lifespan-associated SNPs found by this study versus Mostafavi et al¹⁴. at different nominal p-value thresholds. We used the same binominal regression approach described in text S11. We note that the comparisons in B are likely influenced by several factors that differ between the studies, such as linkage disequilibrium, environmental conditions and covariates, and approaches to genotyping and statistical modeling. We therefore view this comparison as useful for understanding general patterns but note that future work is needed to understand the precise magnitude and specific sources of the observed power differences.