Deep learning and metaheuristic optimization for predicting drug release from nanocarrier systems: a data-driven framework for formulation development

Abstract

Drug-release profiles from nanocarrier systems are governed by nonlinear and multiphasic physicochemical interactions that are difficult to predict using conventional kinetic models. This study presents a data-driven predictive framework that integrates deep neural networks (DNNs) with three bio-inspired metaheuristic optimization algorithms: the Transit Search Optimization Algorithm (TSA), Crocodile Hunting Strategy (CHS), and Osprey Optimization Algorithm (OOA). The models were developed and evaluated using a dataset of 403 nanocarrier formulations compiled from 237 peer-reviewed studies. Predictive performance was assessed using RMSE, MSE, R 2 , Nash-Sutcliffe Efficiency (NSE), and GR125, followed by multi-criteria ranking with TOPSIS and statistical validation using the Kruskal–Wallis and Friedman tests. Among the evaluated models, the OOA-optimized DNN achieved the best overall performance, reaching a minimum RMSE of approximately 1.01 and an R 2 of approximately 0.99 while exhibiting stable convergence and statistically significant improvements over the competing optimization strategies. The proposed framework provides a robust in-silico tool for modeling complex drug-release behavior, reducing experimental screening requirements, and supporting data-driven formulation optimization, Quality by Design (QbD), and Model-Informed Drug Development (MIDD) for nanocarrier systems.