Abstract
OBJECTIVE: This study aimed to investigate the predictive potential of specific plasma proteins for the onset of anxiety and depression.
METHODS: Data from the UK Biobank included individuals diagnosed with depression, anxiety, or both conditions, as well as baseline proteomic profiles. Cox proportional-hazards regression models were utilized to assess the associations between protein levels and disease. Essential biological processes underlying disease mechanisms were determined. A machine learning framework combining LightGBM with sequential forward selection (SFS) was applied to develop optimal predictive protein and visualized via SHapley Additive exPlanations (SHAP) plots. Receiver operating characteristic (ROC) analyses were performed to assess the predictive accuracy.
RESULTS: After excluding participants with self-reported or baseline psychiatric conditions, the three cohorts comprised 48,072, 50,555, and 46,762 participants. GDF15 (depression: hazard ratio (HR) = 1.63, P = 3.21 × 10-74; anxiety: HR = 1.45, P = 9.49 × 10-38; co-occurrence: HR = 1.52, P = 1.85 × 10-14), PLAUR (depression: HR = 2.27, P = 1.07 × 10-44; anxiety: HR = 1.94, P = 3.56 × 10-33; co-occurrence: HR = 2.11, P = 1.92 × 10-11), and TNFRSF10B (depression: HR = 1.35, P = 1.07 × 10-39; anxiety: HR = 1.30, P = 5.11 × 10-29; co-occurrence: HR = 1.34, P = 3.11 × 10-11) were strongly associated with both psychiatric disorders. When combined with demographic indicators, PIGR (AUC = 0.626), a panel of 16 proteins (AUC = 0.617), and PLAUR (AUC = 0.588) demonstrated clinically meaningful predictive value for depression, anxiety, and the co-occurrence of both disorders, respectively.
CONCLUSIONS: This study identifies plasma proteomic alterations associated with the onset of depression and anxiety, highlighting their potential for advancing personalized mental health care.