Schizophrenia, a complex neuropsychiatric disorder, frequently experiences a high rate of misdiagnosis due to subjective symptom assessment. Consequently, there is an urgent need for innovative and objective diagnostic tools. In this study, we used cutting-edge extracellular vesicles' (EVs) proteome profiling and XGBoost-based machine learning to develop new markers and personalized discrimination scores for schizophrenia diagnosis and prediction of treatment response. We analysed plasma and plasma-derived EVs from 343 participants, including 100 individuals with chronic schizophrenia, 34 first-episode and drug-naïve patients, 35 individuals with bipolar disorder, 25 individuals with major depressive disorder and 149 age- and sex-matched healthy controls. Our innovative approach uncovered EVs-based complement changes in patients, specific to their disease-type and status. The EV-based biomarkers outperformed their plasma counterparts, accurately distinguishing schizophrenia individuals from healthy controls with an area under curve (AUC) of 0.895, 83.5% accuracy, 85.3% sensitivity and 82.0% specificity. Moreover, they effectively differentiated schizophrenia from bipolar disorder and major depressive disorder, with AUCs of 0.966 and 0.893, respectively. The personalized discrimination scores provided a personalized diagnostic index for schizophrenia and exhibited a significant association with patients' antipsychotic treatment response in the follow-up cohort. Overall, our study represents a significant advancement in the field of neuropsychiatric disorders, demonstrating the potential of EV-based biomarkers in guiding personalized diagnosis and treatment of schizophrenia.
Keywords: antipsychotic response; extracellular vesicle; machine learning; proteomics; schizophrenia.
© The Author(s) 2023. Published by Oxford University Press on behalf of the Guarantors of Brain.