Identification of Blood Biomarkers for Bipolar Disorder using SERS and Explainable Machine Learning

The differentiation of bipolar disorder (BD) from major depressive disorder (MDD) represents a persistent challenge in psychiatry, leading to high rates of misdiagnosis and adverse patient outcomes. Objective biomarkers are critically needed, but their discovery is hampered by limited understanding of the disease etiology. Robust unbiased, hypothesis-free, and untargeted discovery methods combined with machine learning could prove valuable in providing diagnostic utility. In this study, a novel surface-enhanced Raman spectroscopy (SERS)-gold nanoparticle (AuNP)-cucurbit[5]uril (CB[5]) approach for biomarker discovery was applied for the first time in a clinical setting in combination with an explainable AI driven approach to distinguish BD from MDD. Dried blood spot (DBS) samples were collected from 130 patients with BD and MDD and analysed using SERS and AuNP-CB[5] aggregates. An XGBoost machine learning model was trained to differentiate the groups, with the Context Representative Interpretable Model Explanations (CRIME) method used to identify the most predictive spectral features. These features were subsequently correlated with metabolites measured using targeted mass spectrometry to identify underlying biomarkers. The model distinguished BD from MDD with an AUC of 0.66 (P = 0.003). The predictions were driven by two primary spectral peak regions identified through correlation analysis. The most important peak (1450 cm-¹) corresponded to hypoxanthine, which was significantly upregulated in the BD group. The link between hypoxanthine and the SERS signal was supported by experimental confirmation of correlated spectral peaks, validating the method’s specificity. This study validates a comprehensive DBS-SERS-AuNP-CB[5] clinical discovery pipeline, establishing a powerful, unbiased framework that opens new avenues for developing point-of-care diagnostics for complex neuropsychiatric disorders and other conditions.