%0 Journal Article %T Predicting Antidepressant Treatment Response Using Machine Learning: A Multimodal Analysis of Clinical and Genetic Data %A Rocco de Filippis %A Abdullah Al Foysal %J Open Access Library Journal %V 12 %N 8 %P 1-26 %@ 2333-9721 %D 2025 %I Open Access Library %R 10.4236/oalib.1113958 %X Accurately predicting individual responses to antidepressant treatment is a critical step toward achieving personalized psychiatry and minimizing the traditional trial-and-error approach in clinical practice. This study applies a comprehensive machine learning framework to predict antidepressant treatment outcomes by integrating both clinical and genetic data. Four supervised learning models were developed and evaluated: Random Forest, XGBoost, Support Vector Machine (SVM), and Logistic Regression. The dataset consisted of balanced groups of responders and non-responders, incorporating key clinical variables such as age, body mass index (BMI), baseline depression severity measured by HAMD scores, illness duration, sleep quality, early life stress, and anxiety comorbidity, along with genetic polymorphisms including the 5HTTLPR variant and other serotonin-related markers. Extensive data preprocessing, feature engineering, and hyperparameter tuning using GridSearchCV with five-fold cross-validation were employed to ensure model robustness and reliability. Model evaluation was based on multiple performance metrics, including accuracy, precision, recall, F1 score, and ROC AUC, supported by confusion matrices and visualizations of ROC and precision-recall curves. Feature importance was systematically analyzed using Random Forest rankings, Logistic Regression coefficients, and SHAP (SHapley Additive exPlanations) values to provide model interpretability and clinical insight. Among the evaluated models, Random Forest and Logistic Regression demonstrated the most balanced predictive capabilities. Clinical features, particularly baseline depression severity, age, and early life stress, emerged as the most influential predictors, while the genetic marker 5HTTLPR also showed a significant contribution to treatment response classification. The study further refined clinical applicability through threshold optimization, enhancing recall performance to prioritize responder detection. These findings highlight the potential of machine learning to support personalized treatment strategies in psychiatric care. 
%K Antidepressant Response %K Machine Learning %K Clinical Data %K Genetic Biomarkers %K Precision Psychiatry %K Random Forest %K Logistic Regression %K SHAP %K Threshold Optimization %U http://www.oalib.com/paper/6868325