Interpretable machine learning with SHAP analysis identifies redox-modulating dietary antioxidants for predicting...

Aging is a complex biological process characterized by progressive functional decline across multiple physiological systems, and biological age provides a more accurate reflection of an individual's aging status than chronological age. Dietary antioxidants represent key modifiable targets for delaying aging; however, existing studies lack systematic screening and precise identification of core antioxidant components, and traditional statistical methods have inherent limitations in handling complex associations within high-dimensional dietary data. This study aimed to develop an interpretable machine learning model for predicting accelerated biological aging, and to identify core dietary antioxidants closely associated with accelerated aging risk.

This study analyzed data from the U.S. National Health and Nutrition Examination Survey (NHANES) 2007-2010 and 2017-2018 cycles, with a final total of 8125 participants included in the analysis. Accelerated biological aging, the primary outcome, was defined using two gold-standard metrics: the Klemera-Doubal Method (KDM) and Phenotypic Age (PhenoAge). Eight machine learning models were constructed, and their performance was evaluated via 10-fold cross-validation. The SHapley Additive Explanations (SHAP) framework was applied to quantify the predictive contribution of each feature. Multivariable logistic regression, restricted cubic spline models, subgroup analyses, and interaction tests were performed to validate the robustness of the observed associations and characterize their dose-response relationships.

The XGBoost model exhibited the optimal predictive performance for accelerated aging defined by both metrics, with an area under the receiver operating characteristic curve (AUC) of 0.931 (KDM) and 0.906 (PhenoAge) in 10-fold cross-validation. SHAP analysis consistently identified daidzein, apigenin, magnesium, zinc, and vitamin E as the core dietary antioxidants with the highest predictive contribution to accelerated aging risk. Dose-response analyses revealed significant inverse linear and nonlinear associations between higher intake of these five antioxidants and accelerated aging risk, and these associations were consistently observed across sex and age subgroups.

The interpretable machine learning framework based on dietary antioxidant profiles can robustly predict accelerated biological aging. XGBoost demonstrated the best performance in predicting accelerated aging risk. Daidzein, apigenin, magnesium, zinc, and vitamin E are core dietary antioxidants closely associated with a reduced risk of accelerated biological aging. This study provides analytical reference at the large population level for the development of nutritional intervention strategies targeting healthy aging. Given the cross-sectional design of the study, our findings only reflect predictive associations rather than causal relationships, and warrant further validation in prospective cohort studies and randomized controlled trials.