Liquid\nBiopsy-Based Detection and Response Prediction\nfor Depression

Proactively predicting\nantidepressant treatment response before\nmedication failures is crucial, as it reduces unsuccessful attempts\nand facilitates the development of personalized therapeutic strategies,\nultimately enhancing treatment efficacy. The current decision-making\nprocess, which heavily depends on subjective indicators, underscores\nthe need for an objective, indicator-based approach. This study developed\na method for detecting depression and predicting treatment response\nthrough deep learning-based spectroscopic analysis of extracellular\nvesicles (EVs) from plasma. EVs were isolated from the plasma of both\nnondepressed and depressed groups, followed by Raman signal acquisition,\nwhich was used for AI algorithm development. The algorithm successfully\ndistinguished depression patients from healthy individuals and those\nwith panic disorder, achieving an AUC accuracy of 0.95. This demonstrates\nthe model’s capability to selectively diagnose depression within\na nondepressed group, including those with other mental health disorders.\nFurthermore, the algorithm identified depression-diagnosed patients\nlikely to respond to antidepressants, classifying responders and nonresponders\nwith an AUC accuracy of 0.91. To establish a diagnostic foundation,\nthe algorithm applied explainable AI (XAI), enabling personalized\nmedicine for companion diagnostics and highlighting its potential\nfor the development of liquid biopsy-based mental disorder diagnosis.