Machine Learning identifies conserved traits that influence lifespan and healthspan responses to dietary restriction.

Dietary restriction (DR) is the most robust means to extend lifespan and healthspan across species, but factors such as genetic variation affect how an individual will respond to DR. Additionally, it is unclear how cumulative variations in metabolism and the metabolome influence longevity and health. We utilized metabolomic, phenotypic, and genome-wide data from Drosophila Genetic Reference Panel strains raised under ad libitum and DR conditions to identify factors which influence longevity and health in response to dietary modulation. We found multiple intra-dataset correlations (e.g., metabolites with metabolites) but few inter-dataset correlations (e.g., metabolites with health-related phenotypes). Through random forest modeling across all traits and Mendelian Randomization, we found key translatable traits that influence lifespan or healthspan determination and validated the role of multiple metabolites in regulating lifespan. Through these approaches, we utilized data from flies and humans to elucidate potential therapeutic pathways and metabolomic targets for diet response, lifespan, and healthspan.