Predictive modeling approach using machine learning-integrated design of experiments in quality by design for...

This study aimed to explore the potential of a Machine learning (ML)-integrated Quality by design (QbD) process to formulate resveratrol (RES)-loaded polymeric nanoparticles (RES-PNPs) for potential use in transdermal drug delivery. The RES-PNPs were prepared using the nanoprecipitation method. The critical material attributes (CMAs) and critical process parameters (CPPs) were studied, focusing on particle size (PS), polydispersity index (PDI), zeta potential (ZP), and % encapsulation efficiency (%EE), which were identified as critical quality attributes (CQAs). The regression models from the traditional QbD and the algorithm models from ML-integrated QbD were used to generate prediction models, and their efficiencies were compared. The results showed that the validation accuracy of ML-integrated QbD was higher than that of traditional QbD, indicated by lower root mean squared error (RMSE) and higher R². However, the testing accuracy of traditional QbD significantly decreased due to an underfitting model. The optimal RES-PNPs were selected from a design space comprising 0.1 % polyacrylic acid, 0.5 % gelatin, and 1.11 % Poloxamer 407, with a sonication frequency of 21.43 Hz for 5.02 min. The results revealed PS of 72.85 ± 10.21 nm, PDI of 0.30 ± 0.09, ZP of -18.11 ± 8.56 mV, and % encapsulation efficiency of 79.25 ± 3.22 %. Preparing optimal RES-PNPs from ML-integrated QbD demonstrated high skin permeation, cell viability, antioxidant activity, and stability over one month at 4-25 °C without light exposure. Therefore, ML-integrated QbD was a promising tool that could be used to drive smart pharmaceutical processes in the future.