Precision nutrition in epigenetic aging: SHAP-optimized machine learning identifies omega-3 constituent-specific associations with aging biomarkers.
Journal:
Biogerontology
Published Date:
Jul 24, 2025
Abstract
This cross-sectional investigation seeks to examine the association between dietary omega-3 fatty acids (including α-linolenic acid [ALA], eicosapentaenoic acid [EPA], and docosahexaenoic acid [DHA]) and biomarkers of cellular aging, specifically DNA methylation age (Horvathage) and telomere length (Horvathtelo), in older adults. Our analysis leveraged nationally representative data from 2,136 participants aged ≥ 50 years in the 1999-2002 NHANES cycles. Multivariable linear regression models with survey weights were constructed to evaluate dose-response relationships, complemented by restricted cubic splines (RCS) for nonlinearity detection. Machine learning optimization included comparative evaluation of nine algorithms through five-fold cross-validation, with model interpretability enhanced via SHapley Additive exPlanations (SHAP) analysis. Higher omega-3 intake (Tertile 3 [T3] vs Tertile 1 [T1]) demonstrated inverse associations with HorvathAge (β = -1.07), particularly for ALA intake (T3 ≥ 1.512 g/d: β = -1.11). Contrastingly, moderate-to-high omega-3 intake (T2 ≥ 0.917 g/d: β = 0.04; T3: β = 0.04) and individual components (ALA_T3: β = 0.04; DHA_T3 ≥ 0.041 g/d: β = 0.05; EPA_T3 ≥ 0.011 g/d: β = 0.03) exhibited positive correlations with HorvathTelo. RCS modeling revealed distinct patterns: linear inverse correlation for HorvathAge versus nonlinear J-shaped association with Horvathtelo. Among ML models, Linear Support Vector Machines achieved superior predictive performance. SHAP feature importance analysis consistently ranked omega-3 composite measures highest, followed by constituent components (ALA > DHA > EPA). Our findings suggest a potential dual role of omega-3 in biological aging modulation: higher intake associates with decelerated epigenetic aging while maintaining telomere length homeostasis. These observations underscore the importance of considering both composite measures and individual components in nutritional gerontology research.
