Machine Learning-Enabled Metallomics Reveals Geographic Exposomic Signatures in a Large Brazilian Cohort.

Human exposure to environmental metals and metalloids is shaped by complex interactions among geography, sociodemographic characteristics, lifestyle, and environmental conditions. Although human biomonitoring provides a powerful framework to assess internal exposure, large-scale studies capable of resolving geographically structured exposomic patterns remain scarce. This study aimed to evaluate whether metallomics profiles combined with machine learning can be used to identify Geographic ExposOmic Signatures (GExOS), defined as geographically structured internal exposure patterns derived from integrated biomonitoring data. We profiled the plasma metallome of 9949 adults participating in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Plasma concentrations of 15 metals and metalloids were quantified ICP-MS. Unsupervised multivariate analyses were used to explore global metallomics patterns, followed by supervised machine learning models (XGBoost) to assess geographic classification performance and identify the elements driving regional differentiation. Distinct plasma metallomics profiles were observed according to sex, age, education, income, smoking, and alcohol consumption. Unsupervised analyses revealed structured but continuous regional exposure gradients. Machine learning models demonstrated robust geographic classification performance, with accuracy, sensitivity, and specificity consistently exceeding 80%. Rubidium emerged as a major determinant of geographic discrimination, highlighting its role in shaping spatially structured internal exposure patterns and defining GExOS. These findings demonstrate that the integration of metallomics and machine learning enables the identification of GExOS, providing a scalable and biologically grounded framework to resolve spatial heterogeneity in the human exposome. Metallomics-derived GExOS encode interpretable and predictive exposure signatures with potential applications in environmental surveillance, population health risk assessment, exposure mapping, nutritional epidemiology, and forensic and regulatory sciences.