An integrated evolutionary optimization and machine learning framework for predicting pharmacokinetics of nanocarrier-encapsulated drugs.

This study developed a machine learning framework for predicting the pharmacokinetic behavior of nanocarrier systems using integrated molecular, physicochemical, and formulation descriptors. A Multilayer Perceptron Regressor (MLPR) was employed as the baseline predictive model, while evolutionary optimization algorithms were used to optimize the neural-network architecture and training parameters. The framework was evaluated for predicting three pharmacokinetic endpoints: maximum concentration (Cmax), time to maximum concentration (Tmax), and area under the concentration-time curve (AUC). The combined model produced 20%-30% reductions in RMSE for Cmax and AUC and kept R 2 values above 0.85, while Tmax remained comparatively more challenging to model than Cmax and AUC, reflecting the stronger influence of dynamic absorption and release-related processes on temporal pharmacokinetic behavior. The results demonstrate that optimizer-guided neural-network tuning can improve pharmacokinetic prediction from heterogeneous nanocarrier datasets. The proposed framework provides an approach for improving the prediction of Cmax, Tmax, and AUC while explicitly considering predictive uncertainty and computational trade-offs.