Estimating Genetic Variance in Life-Span Response to Diet: Insights From Statistical Simulation.

Abstract

Several studies demonstrate genetic variation in response to dietary restriction (DR) by replicating treatments across isogenic lines/strains from genetic reference panels. These studies typically quantify the response to DR as an effect size, estimated for each strain separately (eg, the difference in mean life span between groups). Such "no-pooling" analyses are expected to systematically overestimate variation in response DR, principally by overlooking sampling variance. In contrast, "partial-pooling" analyses using mixed-effects models are less prone to this bias. I demonstrate these issues using simulations, which also show that partial-pooling analyses can improve replicability among studies. Regardless of the analyses used, estimates of among-strain variation will have low precision when sample sizes are small. A worked example using survival data in mice is given. Life-span studies using genetic reference panels always have to trade-off within- and among-strain replication owing to logistical challenges. The simulation presented can also be used to help design such studies through power analysis.