Integrative Machine Learning of Genetic and Lifestyle Factors for Personalized Skin Health

To develop an AI framework that combines genetic, phenotypic, and lifestyle data for profiling skin-health patterns and generating hypothesis-supporting summaries for potential decision support.

A dataset of 5,254 individuals integrates six genes (FLG, AQP3, MMP-1, MMP-3, SOD2, GPX), six phenotype severities, and 20+ lifestyle factors. Mutation burden and interactions are tested by ANOVA. K-modes clustering identifies four interpretable dermatological profiles within the cohort and is embedded in leakage-free nested cross-validation (train-only selection; test labels from training centroids). Subtypes are predicted from genetics plus lifestyle using an XGBoost (XGB) classifier; explainability uses gain, permutation importance, and SHAP contributions aggregated across outer folds.

Four subtypes are identified. Mutation burden differed across phenotypes (ANOVA, [Formula: see text]). Interactions are observed for AQP3[Formula: see text]Winter[Formula: see text]Dryness, GPX[Formula: see text]Medication[Formula: see text]Pigmentation, and MMP-3[Formula: see text]City Living[Formula: see text]Redness. Nested-CV prediction achieves [Formula: see text] accuracy with macro-F[Formula: see text] and macro-recall [Formula: see text]. This outperformed unimodal baselines and improved generalization across all folds in practice. Drivers are stable across folds and included scrub usage, stress, sleep, low water intake, menopause, and camouflage habits, alongside oxidative-stress and MMP genes.

Integrating genomic susceptibility with modifiable exposures enables robust, interpretable skin-profile prediction and highlights actionable targets for stratified counseling beyond genetic predisposition.