AI-guided prediction of liposomal multi-antioxidant formulations mimicking the human antioxidant network

Agents4Science 2025 Conference Submission211 Authors

Keywords: Antioxidant network, Liposomes, Machine learning–based prediction, Formulation optimization

TL;DR: AI models accurately predicted and optimized the co-encapsulation efficiency of five antioxidant components in liposomal formulations.

Abstract: The antioxidant network, crucial for protecting the body from oxidative stress (comprising vitamin C, vitamin E, coenzyme Q10, glutathione, and alpha-lipoic acid), faces challenges such as low stability and bioavailability despite its efficacy. Liposomes, as promising drug delivery systems capable of encapsulating both hydrophilic and lipophilic compounds, possess the potential to address these issues. This study aims to utilize artificial intelligence (AI) to predict the encapsulation efficiency (EE\%) and recommend optimal formulations for these five antioxidant components when co-encapsulated in a single liposome formulation. We constructed AI models, including Random Forest, XGBoost, and Neural Networks, based on multi-omics and experimental data, confirming that key features like lipid composition, hydrophilic/lipophilic drug characteristics, and cholesterol ratio play significant roles in predicting co-encapsulation efficiency. The AI models predicted optimal liposome compositions and manufacturing conditions for the antioxidant network, and liposomes prepared accordingly showed a high correlation between predicted and actual experimental values. Transmission electron microscopy (TEM), dynamic light scattering (DLS), and zeta potential ($\zeta$-potential) measurements confirmed that the AI-recommended co-encapsulation compositions exhibited excellent morphological characteristics, appropriate particle size, and stable zeta potential. Finally, the actually measured EE\% showed high efficiency consistent with the AI model's predictions, thereby validating the reliability of AI-based predictions. These results demonstrate that an AI-based approach can significantly enhance the efficiency of developing multi-component liposome formulations for the antioxidant network.

Submission Number: 211