Machine learning algorithms for prediction of side effects development in patients with pharmacoresistance to antipsychotics and antidepressants

Rationale. Artificial intelligence and machine learning allow for development of predictive models using pharmacogenetic testing (PGT) data. It helps to predict the development of side effects (SE) in patients treated with antipsychotics (AP) and antidepressants (AD) and provides personalized approach for the treatment of patients with treatment resistance to antipsychotics and antidepressants. Aim: To compare machine learning algorithms for prediction of side effects development in patients with pharmacoresistance (PR) to antipsychotics and antidepressants. Material and Methods. A retrospective study utilized PGT data of 144 patients (72 males and 72 females, mean age 33±8.4 years) with PR to AP and AD, treated on an outpatient basis for the period from 2016 to 2024. PGT assessed CYP2D6, CYP2C19, CYP1A2, and MDR1 (C3435T) gene polymorphisms conducted in medical laboratories in St. Petersburg (MedLab, Invitro). Machine learning algorithms Lasso, Ridge, Extra Tree (ET), k-Nearest Neighbors (KNN), Naive Bayes (NB), Random Forest (RF), and eXtreme Gradient Boosting (XGB) were used to build the predictive model for SE development. Results. RF algorithm demonstrated the best performance as the predictive model in test sample parameters: ROC-AUC 75.5% [59.6; 89.9], sensitivity 72.2% [55.0; 88.9], and specificity 58.3% [33.3; 81.8]. The main predictors included age, sex, CYP2C19, CYP2D6, CYP1A2, MDR1 C3435T genotypes and alleles, smoking, presence of neurological diseases and substance abuse. Conclusion. Random Forest model machine learning algorithm has demonstrated high efficiency in predicting side effects probability in treatment resistant patients to AP and AD. The model can serve as the basis for future research and development of personalized treatment approach for the patients treated with AP and AD, with the possibility of further integration into Medical Decision Support System.