Metabolic Reprogramming Bridges Environmental Exposure and Tinnitus Severity: Evidence from AI-Driven Serum Multiomics

Tinnitus is a prevalent auditory disorder with poorly understood links between environmental exposure and disease, complicating objective diagnosis and intervention. In a graded observational cohort with documented occupational noise histories, we integrated serum metabolomics, lipidomics, and immunophenotyping with interpretable machine learning. Mediation analysis suggested that a panel of metabolites, including gamma-aminobutyric acid (GABA), fumaric acid, and sphingolipids, accounted for a substantial proportion of the association between noise exposure and tinnitus severity (indirect effect = 92%, p < 0.001). Lipidomic profiling indicated early sphingolipid perturbations, followed by patterns consistent with T helper 1 (Th1)–skewed immune activation and lower circulating GABA, outlining a putative “Exposure–Metabolism–Immunity” cascade. A CatBoost predictive model trained on these features stratified tinnitus severity with 84.8% accuracy and a mean absolute error of 0.174. In summary, our findings identify specific biomarker associations that link occupational noise exposure to the metabolic and immune signatures of tinnitus and generate hypotheses implicating GABAergic tone and sphingolipid metabolism as candidate pathways for future mechanistic and interventional studies.