A machine learning-derived dietary pattern for aging

Abstract

Diet is a modifiable determinant of aging. We develop and validate the Machine-learning YouTHful (MYTH) Diet, a dietary pattern associated with reduced aging-related mortality. Using data from 191,689 participants in the UK Biobank, we conducted a food-wide association analysis and identified 18 food groups significantly associated with aging-related mortality. A Light Gradient Boosting Machine (LightGBM) model was used to rank food importance, leading to the construction of a 10-component MYTH Diet score (range: 0–10). Higher MYTH scores were consistently associated with reduced aging-related mortality in both internal (Quartile 4 vs. 1: hazard ratio [HR] = 0.79; 95% CI: 0.75–0.84) and external (Q4 vs. Q1: HR = 0.68; 95% CI: 0.58–0.80) validation cohorts. Multi-omics analyses revealed that the diet’s protective effects were partly mediated through proteomic, metabolic, and inflammatory pathways, with mediators including TNFRSF4, the proportion of polyunsaturated fatty acids (PUFA%), and lipid-related metabolites such as medium very-low-density lipoprotein phospholipids (M-VLDL-PL). Higher MYTH scores were also linked to slower biological aging in the lungs, liver, pancreas, as well as lower risks for 15 aging-related diseases. These findings suggest that the MYTH Diet may offer a biologically informed, scalable framework for developing personalized nutrition strategies aimed at supporting healthy aging and longevity.