Integrating clinical and biochemical data from healthy participants to predict pharmacokinetic response to brivaracetam via in silico modeling

Purpose Interindividual variability in drug exposure remains a major challenge in clinical pharmacology and precision therapeutics, limiting the prediction of individual pharmacokinetic (PK) responses even for compounds with relatively stable profiles. Brivaracetam (BRV), a highly selective synaptic vesicle protein 2A (SV2A) ligand approved for focal-onset seizures, exhibits relatively low PK variability; however, clinically relevant differences in systemic exposure persist. This study aimed to develop and compare machine learning–based predictive models to identify multidimensional determinants of BRV exposure using baseline clinical and biochemical variables. Methods Data were obtained from a bioequivalence study involving healthy participants receiving a single oral dose of BRV (100 mg). Plasma concentrations were quantified using validated LC–MS/MS methods, and AUC 0-t values were estimated through noncompartmental analysis. Fifty-one baseline predictors were evaluated within a supervised machine learning framework. Support Vector Regression (SVR) and Multilayer Perceptron (MLP) algorithms were implemented using nested cross-validation with iterative feature selection based on permutation importance. Predictive performance was assessed using the coefficient of determination (R 2 ). Results Both models demonstrated strong predictive capability, explaining over 80% of interindividual variability in BRV exposure (SVR: R 2 = 0.879 ± 0.050; MLP: R 2 = 0.819 ± 0.111). Feature selection generated a parsimonious subset of predictors primarily involving hematological, cardiovascular, and metabolic markers. Conclusions Machine learning approaches effectively integrated multidimensional clinical information to capture determinants of BRV PK, supporting future applications in individualized exposure prediction, early-phase studies, and precision pharmacology.