Machine learning and SHAP-based interpretation of dietary antioxidants' impact on comorbid MDD and suicidal ideation

To develop and validate a machine learning model for predicting major depressive disorder (MDD) with suicidal ideation (SI) by incorporating dietary antioxidants, and to elucidate the specific contribution of these antioxidants in the prediction.

Data were obtained from NHANES 2007-2010. Dietary antioxidants, including vitamins, minerals, and polyphenols, were the primary predictors. Demographic, lifestyle, and health-related variables were also included. Collinear variables were removed, class imbalance was corrected, and data were normalized prior to modeling. Twelve machine learning algorithms were compared using a systematic benchmarking protocol within the sklearn framework: Random Forest (RF), LightGBM, AdaBoost, XGBoost, Extra Trees, CatBoost, Gradient Boosting Decision Trees, Support Vector Machine, Decision Tree, Gaussian Naïve Bayes, Stochastic Gradient Descent, and K-Nearest Neighbors. SHapley Additive exPlanation (SHAP) values were calculated to interpret feature contributions within the best-performing model.

A total of 9306 participants were included, of whom 247 with MDD comorbid with SI. After preprocessing, 31 dietary antioxidants and 9 demographic and health-related variables were retained for modeling. The RF classifier achieved the highest performance with an accuracy of 83%, precision of 0.825, recall of 0.838, area under the ROC curve of 0.927, and F1 score of 0.831. SHAP analysis identified vitamin C, kaempferol, myricetin, peonidin, luteolin, eriodictyol, hesperetin, pelargonidin, and zinc as the most influential predictors.

The RF model exhibited superior predictive capability for comorbid MDD and SI. SHAP analysis highlighted the critical roles of dietary antioxidants, particularly vitamin C and kaempferol, in predicting these outcomes.