Supervised machine learning for predicting drug release from acetalated dextran nanofibers

Abstract

Electrospun drug-loaded polymeric nanofibers can improve the efficacy of therapeutics for a variety of implications. By design, these biomaterial platforms can enhance drug bioavailability and site-specific delivery while reducing off-target toxicities when compared to other conventional formulations. By incorporating biocompatible and biodegradable polymers with tunable degradation rates, such as acetalated dextran (Ace-DEX), drug-loaded nanofibers can enhance the safety and efficacy of treatment regimens while improving patient compliance through controlled release. Despite these benefits, clinical translation of electrospun formulations is challenged by labor-intensive in vitro studies for ensuring that release kinetics are accurately characterized and reproducible. In this study, we report a novel workflow for assessing in vitro drug release from Ace-DEX nanofibers using machine learning (ML) and develop a predictive model to streamline this rate-limiting step. The developed Gaussian process regression (GPR) model was trained, validated, and optimized using in vitro release profiles from thirty electrospun Ace-DEX scaffolds. The results of GPR model simulations reveal consistent performance across all Ace-DEX formulations considered in this study while also demonstrating a drug-agnostic approach to predict fractional drug release over time.